Search

In recent years, the socio-economic development in the Tibet region of China has experienced substantial growth. However, transportation increasingly strains the region's fragile ecological environment. Most studies overlook the accurate measurement and analysis of factors influencing traffic carbon emissions in Tibet due to data scarcity. To address this, this paper applies an improved traffic carbon emissions model, using transportation turnover data to estimate emissions in Tibet from 2008 to 2020. Simultaneously, the estimated traffic carbon emissions in Tibet served as the predicted variable, and various machine learning algorithms, including Radial Basis Function Support Vector Machine (RBF-SVM), eXtreme Gradient Boosting (XGBoost), Random Forest, and Gradient Boosting Decision Tree (GBDT) are employed to conduct an initial comparison of the constructed prediction models using three-fold cross-validation and multiple evaluation metrics. The best-performing model undergoes further optimization using Grid Search (GS) and Real-coded Genetic Algorithm (RGA). Finally, the central difference method and Local Interpretable Model-Agnostic Explanation (LIME) algorithm are used for local sensitivity and interpretability analyses on twelve core variables. The results assess each variable's contribution to the model's output, enabling a comprehensive analysis of their impact on Tibet's traffic carbon emissions. The findings demonstrate a significant upward trend in Tibet's traffic carbon emissions, with road transportation and civil aviation being the main contributors. The RBF-SVM algorithm is most suitable for predicting traffic carbon emissions in this region. After GS optimization, the model's R2 value exceeded 0.99, indicating high predictive accuracy and stability. Key factors influencing traffic carbon emissions in Tibet include civilian vehicle numbers, transportation land-use area, transportation output value, urban green coverage areas, per capita GDP, and built-up area. This paper provides a systematic framework and empirical support for measuring, predicting, and analyzing factors influencing traffic carbon emissions in Tibet. It employs innovative measurement methods, optimized machine learning models, and detailed sensitivity and interpretability analyses. The results can guide regional carbon reduction targets and promote green sustainable development. [ABSTRACT FROM AUTHOR]

Although live streaming is indispensable, live-streaming e-business requires accurate and timely sales-volume prediction to ensure a healthy supply–demand balance for companies. Practically, because various factors can significantly impact sales results, the development of a powerful, interpretable model is crucial for accurate sales prediction. In this study, we propose SaleNet, a deep-learning model designed for sales-volume prediction. Our model achieved correct prediction results on our private, real operating data. The mean absolute percentage error (MAPE) of our model's performance fell as low as 11.47% for a + 1.5-days forecast. Even for a 1-week forecast (+ 6 days), the MAPE was only 19.79%, meeting actual business needs and practical requirements. Notably, our model demonstrated robust interpretability, as evidenced by the feature contribution results which are consistent with prevailing research findings and industry expertise. Our findings provided a theoretical foundation for predicting shopping behavior in live-broadcast e-commerce and offered valuable insights for designing live-broadcast content and optimizing the user experience. [ABSTRACT FROM AUTHOR]

Applying deep learning to high-precision ionospheric parameter prediction is a significant and growing field within the realm of space weather research. This paper proposes an improved model, Mixed Convolutional Neural Network (CNN)—Bidirectional Long Short-Term Memory (BiLSTM), for predicting the Total Electron Content (TEC) in China. This model was trained using the longest available Global Ionospheric Maps (GIM)-TEC from 1998 to 2023 in China, and underwent an interpretability analysis and accuracy evaluation. The results indicate that historical TEC maps play the most critical role, followed by Kp, ap, AE, F10.7, and time factor. The contributions of Dst and Disturbance Index (DI) to improving accuracy are relatively small but still essential. In long-term predictions, the contributions of the geomagnetic index, solar activity index, and time factor are higher. In addition, the model performs well in short-term predictions, accurately capturing the occurrence, evolution, and classification of ionospheric storms. However, as the predicted length increases, the accuracy gradually decreases, and some erroneous predictions may occur. The northeast region exhibits lower accuracy but a higher F1 score, which may be attributed to the frequency of ionospheric storm occurrences in different locations. Overall, the model effectively predicts the trends and evolution processes of ionospheric storms. [ABSTRACT FROM AUTHOR]

Abstract Although live streaming is indispensable, live-streaming e-business requires accurate and timely sales-volume prediction to ensure a healthy supply–demand balance for companies. Practically, because various factors can significantly impact sales results, the development of a powerful, interpretable model is crucial for accurate sales prediction. In this study, we propose SaleNet, a deep-learning model designed for sales-volume prediction. Our model achieved correct prediction results on our private, real operating data. The mean absolute percentage error (MAPE) of our model’s performance fell as low as 11.47% for a + 1.5-days forecast. Even for a 1-week forecast (+ 6 days), the MAPE was only 19.79%, meeting actual business needs and practical requirements. Notably, our model demonstrated robust interpretability, as evidenced by the feature contribution results which are consistent with prevailing research findings and industry expertise. Our findings provided a theoretical foundation for predicting shopping behavior in live-broadcast e-commerce and offered valuable insights for designing live-broadcast content and optimizing the user experience.

Abstract The preservation and transmission of traditional villages is crucial to the prosperity and development of ethnic cultures. However, current traditional village surveys usually require a large number of experts and scholars to conduct field research, which is an expensive and time-consuming method, especially for large-scale tasks. Therefore, this study proposes an automatic classification method based on deep learning (DL) for the identification of traditional village heritage value elements (TVHVE). The study evaluates four selected convolutional neural network (CNN) frames using traditional villages in Hubei Province as a sample dataset. The results show that Residual Network152 (ResNet152) is the most suitable CNN frame for identifying TVHVE in Hubei. The stability and consistency of various TVHVE present in the ResNet152 model were evaluated using Area Under Curve (AUC) and Precision Recall Curve (PRC), which indicated satisfactory prediction performance for most elements, except for specific elements such as tombstones and stone carvings, which showed lower accuracy. In addition, the study sheds light on the areas of concern of the model with respect to different TVHVE images and elucidates the reasons behind the confusion between elements through semantic clustering based on image classification and interpretability analysis using the Gradient-Weighted Class Activation Mapping (Grad-CAM) heat map. By using an automated classification method based on DL, this study significantly reduces the cost and effort associated with traditional surveys. At the same time, insight into areas of concern and confusion in the model improves guidance for conservation efforts and provides valuable references for subsequent research.

The application of machine learning (ML) is becoming widespread and playing an important role in the property prediction and design of battery. Average voltage and specific capacity are two important indicators of electrode materials, which determine suitability of electrode materials in batteries. In this study, 4351 data are collected from Material Project and 324 features are extracted from Material Project or calculated by Matminer. Four ML models are discussed for prediction of average voltage and specific capacity. In order to ensure the prediction accuracy and efficiency of ML models, sequential backward selection (SBS) method is introduced to select optimal feature set, which can reduce redundancy features. The combination of Deep Neural Network model and optimal feature set selected by the SBS method achieves accurate prediction and outperforms Crystal Graph Convolutional Neural Network and Graph‐attention Graph Neural Network. The interpretability analysis provides insight into the relationship between features and two target properties. To cope with the poor prediction performance for metal‐ion battery with scarce data, transfer learning is adopted and an excellent improvement is achieved in the prediction in Na‐ion, Mg‐ion and Ca‐ion electrode material. It can be concluded that ML is an effective approach to battery property prediction and design. [ABSTRACT FROM AUTHOR]

Accurate understanding of the biological functions of enzymes is vital for various tasks in both pathologies and industrial biotechnology. However, the existing methods are usually not fast enough and lack explanations on the prediction results, which severely limits their real-world applications. Following our previous work, DEEPre , we propose a new interpretable and fast version (ifDEEPre) by designing novel self-guided attention and incorporating biological knowledge learned via large protein language models to accurately predict the commission numbers of enzymes and confirm their functions. Novel self-guided attention is designed to optimize the unique contributions of representations, automatically detecting key protein motifs to provide meaningful interpretations. Representations learned from raw protein sequences are strictly screened to improve the running speed of the framework, 50 times faster than DEEPre while requiring 12.89 times smaller storage space. Large language modules are incorporated to learn physical properties from hundreds of millions of proteins, extending biological knowledge of the whole network. Extensive experiments indicate that ifDEEPre outperforms all the current methods, achieving more than 14.22% larger F1-score on the NEW dataset. Furthermore, the trained ifDEEPre models accurately capture multi-level protein biological patterns and infer evolutionary trends of enzymes by taking only raw sequences without label information. Meanwhile, ifDEEPre predicts the evolutionary relationships between different yeast sub-species, which are highly consistent with the ground truth. Case studies indicate that ifDEEPre can detect key amino acid motifs, which have important implications for designing novel enzymes. A web server running ifDEEPre is available at https://proj.cse.cuhk.edu.hk/aihlab/ifdeepre/ to provide convenient services to the public. Meanwhile, ifDEEPre is freely available on GitHub at https://github.com/ml4bio/ifDEEPre/. [ABSTRACT FROM AUTHOR]

The preservation and transmission of traditional villages is crucial to the prosperity and development of ethnic cultures. However, current traditional village surveys usually require a large number of experts and scholars to conduct field research, which is an expensive and time-consuming method, especially for large-scale tasks. Therefore, this study proposes an automatic classification method based on deep learning (DL) for the identification of traditional village heritage value elements (TVHVE). The study evaluates four selected convolutional neural network (CNN) frames using traditional villages in Hubei Province as a sample dataset. The results show that Residual Network152 (ResNet152) is the most suitable CNN frame for identifying TVHVE in Hubei. The stability and consistency of various TVHVE present in the ResNet152 model were evaluated using Area Under Curve (AUC) and Precision Recall Curve (PRC), which indicated satisfactory prediction performance for most elements, except for specific elements such as tombstones and stone carvings, which showed lower accuracy. In addition, the study sheds light on the areas of concern of the model with respect to different TVHVE images and elucidates the reasons behind the confusion between elements through semantic clustering based on image classification and interpretability analysis using the Gradient-Weighted Class Activation Mapping (Grad-CAM) heat map. By using an automated classification method based on DL, this study significantly reduces the cost and effort associated with traditional surveys. At the same time, insight into areas of concern and confusion in the model improves guidance for conservation efforts and provides valuable references for subsequent research. [ABSTRACT FROM AUTHOR]

Geological landslide disasters significantly threaten the safety of people's lives and property. Landslides are a significant threat in Zayu County, Tibet, resulting in numerous geological disasters, including the 1950 earthquake that caused significant casualties and river blockages. More recent landslides have caused substantial economic losses and infrastructure damage, posing ongoing risks to the local population and their property. Landslide hazard assessment is a critical task in geological disaster prevention and mitigation. This study applied the Inception model to assess landslide hazard in the Zayu area. The Inception model excels at capturing multi-scale features efficiently through its architecture. Fifteen disaster-causing factors were selected as the primary indicators for landslide susceptibility assessment. On this basis, the Inception model was used for landslide susceptibility assessment. Combined with daily precipitation data in the Zayu area, the landslide hazard assessment of the "25 April 2010, heavy rainstorm in Zayu, Tibet" was completed. Back Propagation Neural Network (BPNN), Residual Neural Network (ResNet), Convolutional Neural Network (CNN), and Visual Geometry Group-16 (VGG-16) were introduced for comparison of the fitting effects, and SHapley Additive exPlanations (SHAP) was used for interpretability analysis. The comparative experimental results show that the Inception model performed best in landslide susceptibility assessment and is feasible in practical use. The results also show that the most critical factors in the model were topographic wetness index (TWI), normalized difference water index (NDWI), and road density. This study is significant for assessing landslide hazard in geological landslide disaster prevention and mitigation. It provides a reference for further research and response to similar disasters. [ABSTRACT FROM AUTHOR]

To overcome critical issues in landslide susceptibility modeling, a multifactor landslide susceptibility prediction model based on deep belief networks (DBN) with optimized learning rate control (LRC-DBN) was introduced in this study. The LRC-DBN model was applied to predict landslide susceptibility in Zhidan County, Shaanxi and its performance was compared against random forest, support vector machine and logistic regression models. The results show that the LRC-DBN model achieves a maximum AUC value of 0.941, which demonstrating higher predictive performance. The interpretability of LRC-DBN reveals that the development of loess landslides is more likely in areas with elevations ranging from 894.7 m to 998.5 m, slopes ranging from 49.6° to 64°, terrain undulations spanning 31.7 m to 91.0 m, rock type T3y and terrain humidity ranging from 2.4 to 3.9. The result provides invaluable insights for local landslide prevention endeavors and can assist decision-making in urban planning. [ABSTRACT FROM AUTHOR]

Accurate quantification of soil moisture is essential for understanding water and energy exchanges between the atmosphere and the Earth’s surface, as well as for agricultural applications. Predicting soil moisture content is vital for efficient water management, irrigation scheduling, and drought monitoring. Traditional forecasting methods, such as numerical regression models, often struggle due to various influencing factors and poor observation data quality. In contrast, deep learning algorithms, particularly recurrent and convolutional neural networks, show promise in predicting nonlinear data like soil moisture. This study focuses on shallow groundwater regions, using groundwater levels and meteorological data as features while coupling the Transformer model with other neural network structures. We investigate the potential of attention-based neural networks for soil moisture time series prediction. Our findings demonstrate that the Transformer model achieves an average R2 of 0.523 across different time lags, outperforming the LSTM model with an R2 of 0.485. The introduction of LSTM enhances the Transformer’s stability in handling temporal changes. Additionally, we verified the importance of groundwater for soil moisture prediction. This study introduces new methods for soil moisture prediction and offers new insights and recommendations for the development of artificial intelligence technology for soil moisture prediction.

Considering the wide utilization of multi-component supplementary cementitious materials in recycled aggregate concrete, an intelligent prediction of comprehensive mechanical properties of this sustainable low-carbon concrete was proposed in this study using hybrid machine learning algorithms, rather than conducting tremendous experiments to establish a statistical empirical model previously. The database consisting of 980 datasets was used to train, validate and test the machine learning algorithms with various evaluation indicators. SHapley Additive exPlanations (SHAP) values were used for interpretability analysis of the predictive results of algorithms to quantify the contribution of each input feature. The results show that the hybrid algorithms possess higher prediction accuracy in terms of compressive strength, flexural strength, splitting tensile strength and elastic modulus compared to the single algorithms, among which Sparrow Search Algorithm-Extreme Gradient Boosting (SSA-XGB) stands out as the most promising algorithm model with satisfactory stability and generalization performance. For all mechanical properties, the effective water-binder ratio, cement content, and superplasticizer dosage are extremely important influencing factors. This study provides a systematic assessment framework for predicting the comprehensive mechanical properties of recycled aggregate concrete with supplementary cementitious materials, beneficial for the data-driven intelligent mix proportions design of sustainable low-carbon concrete in construction sector.

The widespread use of encrypted traffic poses challenges to network management and network security. Traditional machine learning-based methods for encrypted traffic classification no longer meet the demands of management and security. The application of deep learning technology in encrypted traffic classification significantly improves the accuracy of models. This study focuses primarily on encrypted traffic classification in the fields of network analysis and network security. To address the shortcomings of existing deep learning-based encrypted traffic classification methods in terms of computational memory consumption and interpretability, we introduce a Parameter-Efficient Fine-Tuning method for efficiently tuning the parameters of an encrypted traffic classification model. Experimentation is conducted on various classification scenarios, including Tor traffic service classification and malicious traffic classification, using multiple public datasets. Fair comparisons are made with state-of-the-art deep learning model architectures. The results indicate that the proposed method significantly reduces the scale of fine-tuning parameters and computational resource usage while achieving performance comparable to that of the existing best models. Furthermore, we interpret the learning mechanism of encrypted traffic representation in the pre-training model by analyzing the parameters and structure of the model. This comparison validates the hypothesis that the model exhibits hierarchical structure, clear organization, and distinct features. [ABSTRACT FROM AUTHOR]

Accurately delineating the connection between short nucleolar RNA (snoRNA) and disease is crucial for advancing disease detection and treatment. While traditional biological experimental methods are effective, they are labor-intensive, costly and lack scalability. With the ongoing progress in computer technology, an increasing number of deep learning techniques are being employed to predict snoRNA–disease associations. Nevertheless, the majority of these methods are black-box models, lacking interpretability and the capability to elucidate the snoRNA–disease association mechanism. In this study, we introduce IGCNSDA, an innovative and interpretable graph convolutional network (GCN) approach tailored for the efficient inference of snoRNA–disease associations. IGCNSDA leverages the GCN framework to extract node feature representations of snoRNAs and diseases from the bipartite snoRNA-disease graph. SnoRNAs with high similarity are more likely to be linked to analogous diseases, and vice versa. To facilitate this process, we introduce a subgraph generation algorithm that effectively groups similar snoRNAs and their associated diseases into cohesive subgraphs. Subsequently, we aggregate information from neighboring nodes within these subgraphs, iteratively updating the embeddings of snoRNAs and diseases. The experimental results demonstrate that IGCNSDA outperforms the most recent, highly relevant methods. Additionally, our interpretability analysis provides compelling evidence that IGCNSDA adeptly captures the underlying similarity between snoRNAs and diseases, thus affording researchers enhanced insights into the snoRNA–disease association mechanism. Furthermore, we present illustrative case studies that demonstrate the utility of IGCNSDA as a valuable tool for efficiently predicting potential snoRNA–disease associations. The dataset and source code for IGCNSDA are openly accessible at: https://github.com/altriavin/IGCNSDA. [ABSTRACT FROM AUTHOR]

In the existing landslide susceptibility prediction (LSP) models, the influences of random errors in landslide conditioning factors on LSP are not considered, instead the original conditioning factors are directly taken as the model inputs, which brings uncertainties to LSP results. This study aims to reveal the influence rules of the different proportional random errors in conditioning factors on the LSP uncertainties, and further explore a method which can effectively reduce the random errors in conditioning factors. The original conditioning factors are firstly used to construct original factors-based LSP models, and then different random errors of 5%, 10%, 15% and 20% are added to these original factors for constructing relevant errors-based LSP models. Secondly, low-pass filter-based LSP models are constructed by eliminating the random errors using low-pass filter method. Thirdly, the Ruijin County of China with 370 landslides and 16 conditioning factors are used as study case. Three typical machine learning models, i.e. multilayer perceptron (MLP), support vector machine (SVM) and random forest (RF), are selected as LSP models. Finally, the LSP uncertainties are discussed and results show that: (1) The low-pass filter can effectively reduce the random errors in conditioning factors to decrease the LSP uncertainties. (2) With the proportions of random errors increasing from 5% to 20%, the LSP uncertainty increases continuously. (3) The original factors-based models are feasible for LSP in the absence of more accurate conditioning factors. (4) The influence degrees of two uncertainty issues, machine learning models and different proportions of random errors, on the LSP modeling are large and basically the same. (5) The Shapley values effectively explain the internal mechanism of machine learning model predicting landslide susceptibility. In conclusion, greater proportion of random errors in conditioning factors results in higher LSP uncertainty, and low-pass filter can effectively reduce these random errors.

The outbreak of COVID-19 has shocked the entire world with its fairly rapid spread, and has challenged different sectors. One of the most effective ways to limit its spread is the early and accurate diagnosing of infected patients. Medical imaging, such as X-ray and computed tomography (CT), combined with the potential of artificial intelligence (AI), plays an essential role in supporting medical personnel in the diagnosis process. Thus, in this article, five different deep learning models (ResNet18, ResNet34, InceptionV3, InceptionResNetV2, and DenseNet161) and their ensemble, using majority voting, have been used to classify COVID-19, pneumoniæ and healthy subjects using chest X-ray images. Multilabel classification was performed to predict multiple pathologies for each patient, if present. Firstly, the interpretability of each of the networks was thoroughly studied using local interpretability methods—occlusion, saliency, input X gradient, guided backpropagation, integrated gradients, and DeepLIFT—and using a global technique—neuron activation profiles. The mean micro F1 score of the models for COVID-19 classifications ranged from 0.66 to 0.875, and was 0.89 for the ensemble of the network models. The qualitative results showed that the ResNets were the most interpretable models. This research demonstrates the importance of using interpretability methods to compare different models before making a decision regarding the best performing model. [ABSTRACT FROM AUTHOR]

Crack detection is an important task in bridge health monitoring, and related detection methods have gradually shifted from traditional manual methods to intelligent approaches with convolutional neural networks (CNNs) in recent years. Due to the opaque process of training and operating CNNs, if the learned features for identifying cracks in the network are not evaluated, it may lead to safety risks. In this study, to evaluate the recognition basis of different crack detection networks; several crack detection CNNs are trained using the same training conditions. Afterwards, several crack images are used to construct a dataset, which are used to interpret and analyze the trained networks and obtain the learned features for identifying cracks. Additionally, a crack identification performance criterion based on interpretability analysis is proposed. Finally, a training framework is introduced based on the issues reflected in the interpretability analysis. [ABSTRACT FROM AUTHOR]

Applying deep learning to high-precision ionospheric parameter prediction is a significant and growing field within the realm of space weather research. This paper proposes an improved model, Mixed Convolutional Neural Network (CNN)—Bidirectional Long Short-Term Memory (BiLSTM), for predicting the Total Electron Content (TEC) in China. This model was trained using the longest available Global Ionospheric Maps (GIM)-TEC from 1998 to 2023 in China, and underwent an interpretability analysis and accuracy evaluation. The results indicate that historical TEC maps play the most critical role, followed by Kp, ap, AE, F10.7, and time factor. The contributions of Dst and Disturbance Index (DI) to improving accuracy are relatively small but still essential. In long-term predictions, the contributions of the geomagnetic index, solar activity index, and time factor are higher. In addition, the model performs well in short-term predictions, accurately capturing the occurrence, evolution, and classification of ionospheric storms. However, as the predicted length increases, the accuracy gradually decreases, and some erroneous predictions may occur. The northeast region exhibits lower accuracy but a higher F1 score, which may be attributed to the frequency of ionospheric storm occurrences in different locations. Overall, the model effectively predicts the trends and evolution processes of ionospheric storms.

Geological landslide disasters significantly threaten the safety of people’s lives and property. Landslides are a significant threat in Zayu County, Tibet, resulting in numerous geological disasters, including the 1950 earthquake that caused significant casualties and river blockages. More recent landslides have caused substantial economic losses and infrastructure damage, posing ongoing risks to the local population and their property. Landslide hazard assessment is a critical task in geological disaster prevention and mitigation. This study applied the Inception model to assess landslide hazard in the Zayu area. The Inception model excels at capturing multi-scale features efficiently through its architecture. Fifteen disaster-causing factors were selected as the primary indicators for landslide susceptibility assessment. On this basis, the Inception model was used for landslide susceptibility assessment. Combined with daily precipitation data in the Zayu area, the landslide hazard assessment of the “25 April 2010, heavy rainstorm in Zayu, Tibet” was completed. Back Propagation Neural Network (BPNN), Residual Neural Network (ResNet), Convolutional Neural Network (CNN), and Visual Geometry Group-16 (VGG-16) were introduced for comparison of the fitting effects, and SHapley Additive exPlanations (SHAP) was used for interpretability analysis. The comparative experimental results show that the Inception model performed best in landslide susceptibility assessment and is feasible in practical use. The results also show that the most critical factors in the model were topographic wetness index (TWI), normalized difference water index (NDWI), and road density. This study is significant for assessing landslide hazard in geological landslide disaster prevention and mitigation. It provides a reference for further research and response to similar disasters.

Loess areas have a unique geological environment, and geological disasters occur there frequently. In this work, the typical loess area Lvliang was used as the study area. Using the historical landslide catalog, 12 influencing factors were chosen by integrating multisource heterogeneous spatiotemporal big data such as remote sensing, ground investigation, and basic geography. Based on frequency ratio (FR) and improved TabNet deep learning technology, landslide susceptibility evaluation and uncertainty analysis were performed. The results showed that the TabNet evaluation model using FR and self-supervised learning performs well and has the highest FR in extremely high-prone areas. Compared with other methods, this method has the highest scores in areas under the curve and susceptibility index distribution and the lowest uncertainty. Moreover, the SHAP method was used for interpretability analysis of the model. Therefore, this study can provide new ideas for landslide susceptibility management.

With the rapid development of deep learning techniques, the applications have become increasingly widespread in various domains. However, traditional deep learning methods are often referred to as “black box” models with low interpretability of their results, posing challenges for their application in certain critical domains. In this study, we propose a comprehensive method for the interpretability analysis of sentiment models. The proposed method encompasses two main aspects: attention-based analysis and external knowledge integration. First, we train the model within sentiment classification and generation tasks to capture attention scores from multiple perspectives. This multi-angle approach reduces bias and provides a more comprehensive understanding of the underlying sentiment. Second, we incorporate an external knowledge base to improve evidence extraction. By leveraging character scores, we retrieve complete sentiment evidence phrases, addressing the challenge of incomplete evidence extraction in Chinese texts. Experimental results on a sentiment interpretability evaluation dataset demonstrate the effectiveness of our method. We observe a notable increase in accuracy by 1.3%, Macro-F1 by 13%, and MAP by 23%. Overall, our approach offers a robust solution for enhancing the interpretability of sentiment models by combining attention-based analysis and the integration of external knowledge.

Drug-target interaction (DTI) is a critical and complex process that plays a vital role in drug discovery and design. In deep learning-based DTI methods, graph neural networks (GNNs) are employed for drug molecule modeling, attention mechanisms are utilized to simulate the interaction between drugs and targets. However, existing methods still face two limitations in these aspects. First, GNN primarily focus on local neighboring nodes, making it difficult to capture the global 3D structure and edge information. Second, the current attention-based methods for modeling drug-target interactions lack interpretability and do not fully utilize the deep representations of drugs and targets. To address the aforementioned issues, we propose an interpretable network architecture called CSDTI. It utilizes a cross-attention mechanism to capture the interaction features between drugs and targets. Meanwhile, we design a drug molecule aggregator to capture high-order dependencies within the drug molecular graph. These features are then utilized simultaneously for downstream tasks. Through rigorous experiments, we have demonstrated that CSDTI outperforms state-of-the-art methods in terms of performance metrics such as AUC, precision, and recall in DTI prediction tasks. Furthermore, the visualization mapping of attention weights indicates that CSDTI can provide chemical insights even without external knowledge. [ABSTRACT FROM AUTHOR]

Optogenetics combines optical and genetic methods to modulate light-controlled gene expression, protein localization, signal transduction and protein interactions to achieve precise control of specific neuronal activity, with the advantages of low tissue damage, high spatial and temporal resolution, and genetic specificity. It provides a cutting-edge approach to establishing a causal relationship between brain activity and behaviors associated with health and disease. Channelrhodopsin (ChR) functions as a photogenic activator for the control of neurons. As a result, ChR and its variants are more widely used in the realization of optogenetics. To enable effective optogenetics, we propose a novel multi-model machine learning framework, i.e., PCSboost, to accurately assist key fragments selection of ChRs segments that realize optogenetics from protein sequence structure and information dataset. We investigate the key regions of the ChR variant protein fragments that impact photocurrent properties of interest and automatically screen important fragments that realize optogenetics. To address the issue of the dataset containing a limited quantity of data but a high feature dimension, we employ principal component analysis (PCA) to reduce the dimensionality of the data and perform feature extraction, followed by the XGBoost model to classify the ChRs based on their kinetics, photocurrent and spectral properties. Simultaneously, we employ the SHAP interpretability analysis to perform an interpretability analysis of the ChR variant protein for pointwise, characteristic similarities to identify key regions of the protein fragment structure that contribute to the regulation of photocurrent intensity, photocurrent wavelength sensitivity and nonkinetic properties. Experimental findings demonstrate that our proposed PCSboost approach can speed up genetic and protein engineering investigations, simplify the screening of important protein fragment sections, and potentially be used to advance research in the areas of optogenetics, genetic engineering and protein engineering. [ABSTRACT FROM AUTHOR]

Background: An effective identification model is crucial to realise the real-time monitoring and early warning of forest fires from surveillance cameras. However, existing models are prone to generate numerous false alarms under the interference of artificial smoke such as industrial smoke and villager cooking smoke, therefore a superior identification model is urgently needed. Aims: In this study, we tested the Transformer-based model FireFormer to predict the risk probability of forest fire from the surveillance images. Methods: FireFormer uses a shifted window self-attention module to extract similarities of divided patches in the image. The similarity in characteristics indicated the probability of forest fires. The GradCAM algorithm was then applied to analyse the interest area of FireFormer model and visualise the contribution of different image patches by calculating gradient reversely. To verify our model, the monitoring data from the high-point camera in Nandan Mountain, Foshan City, was collected and further constructed as a forest fire alarm dataset. Key results: Our results showed that FireFormer achieved a competitive performance (OA: 82.21%, Recall: 86.635% and F 1-score: 74.68%). Conclusions: FireFormer proves to be superior to traditional methods. Implications: FireFormer provides an efficient way to reduce false alarms and avoid heavy manual re-checking work. A Transformer-based model was proposed to identify the forest fire from surveillance camera images. This model proves to be superior in reducing the false alarm rate caused by the interference of different types of smoke, enabling to improve the efficiency of real-time monitoring and early warning of forest fires. [ABSTRACT FROM AUTHOR]

The widespread use of encrypted traffic poses challenges to network management and network security. Traditional machine learning-based methods for encrypted traffic classification no longer meet the demands of management and security. The application of deep learning technology in encrypted traffic classification significantly improves the accuracy of models. This study focuses primarily on encrypted traffic classification in the fields of network analysis and network security. To address the shortcomings of existing deep learning-based encrypted traffic classification methods in terms of computational memory consumption and interpretability, we introduce a Parameter-Efficient Fine-Tuning method for efficiently tuning the parameters of an encrypted traffic classification model. Experimentation is conducted on various classification scenarios, including Tor traffic service classification and malicious traffic classification, using multiple public datasets. Fair comparisons are made with state-of-the-art deep learning model architectures. The results indicate that the proposed method significantly reduces the scale of fine-tuning parameters and computational resource usage while achieving performance comparable to that of the existing best models. Furthermore, we interpret the learning mechanism of encrypted traffic representation in the pre-training model by analyzing the parameters and structure of the model. This comparison validates the hypothesis that the model exhibits hierarchical structure, clear organization, and distinct features.

Introduction: This study aimed to develop novel machine learning models for predicting Alzheimer's disease (AD) and identify key factors for targeted prevention. Methods: We included 1,219, 863, and 482 participants aged 60+ years with only sociodemographic, both sociodemographic and self-reported health, both the former two and blood biomarkers information from Alzheimer's Disease Neuroimaging Initiative (ADNI) database. Machine learning models were constructed for predicting the risk of AD for the above three populations. Model performance was evaluated by discrimination, calibration, and clinical usefulness. SHapley Additive exPlanation (SHAP) was applied to identify key predictors of optimal models. Results: The mean age was 73.49, 74.52, and 74.29 years for the three populations, respectively. Models with sociodemographic information and models with both sociodemographic and self-reported health information showed modest performance. For models with sociodemographic, self-reported health, and blood biomarker information, their overall performance improved substantially, specifically, logistic regression performed best, with an AUC value of 0.818. Blood biomarkers of ptau protein and plasma neurofilament light, age, blood tau protein, and education level were top five significant predictors. In addition, taurine, inosine, xanthine, marital status, and L.Glutamine also showed importance to AD prediction. Conclusion: Interpretable machine learning showed promise in screening high-risk AD individual and could further identify key predictors for targeted prevention. [ABSTRACT FROM AUTHOR]

Despite achieving success in many domains, deep learning models remain mostly black boxes, especially in electroencephalogram (EEG)-related tasks. Meanwhile, understanding the reasons behind model predictions is quite crucial in assessing trust and performance promotion in EEG-related tasks. In this work, we explore the use of representative interpretable models to analyze the learning behavior of convolutional neural networks (CNN) in EEG-based emotion recognition. According to the interpretable analysis, we find that similar features captured by our model and state-of-the-art model are consistent with previous brain science findings. Next, we propose a new model by integrating brain science knowledge with the interpretability analysis results in the learning process. Our knowledge-integrated model achieves better recognition accuracy on standard EEG-based recognition datasets. [ABSTRACT FROM AUTHOR]

Crack detection is an important task in bridge health monitoring, and related detection methods have gradually shifted from traditional manual methods to intelligent approaches with convolutional neural networks (CNNs) in recent years. Due to the opaque process of training and operating CNNs, if the learned features for identifying cracks in the network are not evaluated, it may lead to safety risks. In this study, to evaluate the recognition basis of different crack detection networks; several crack detection CNNs are trained using the same training conditions. Afterwards, several crack images are used to construct a dataset, which are used to interpret and analyze the trained networks and obtain the learned features for identifying cracks. Additionally, a crack identification performance criterion based on interpretability analysis is proposed. Finally, a training framework is introduced based on the issues reflected in the interpretability analysis.

In Natural Language Processing (NLP), deep-learning neural networks have superior performance but pose transparency and explainability barriers, due to their black box nature, and, thus, there is lack of trustworthiness. On the other hand, classical machine learning techniques are intuitive and easy to understand but often cannot perform satisfactorily. Fortunately, many research studies have recently indicated that the newly introduced model, Tsetlin Machine (TM), has reliable performance and, at the same time, enjoys human-level interpretability by nature, which is a promising approach to trade off effectiveness and interpretability. However, nearly all of the related works so far have concentrated on the English language, while research on other languages is relatively scarce. So, we propose a novel method, based on the TM model, in which the learning process is transparent and easily-understandable for Chinese NLP tasks. Our model can learn semantic information in the Chinese language by clauses. For evaluation, we conducted experiments in two domains, namely sentiment analysis and spam review detection. The experimental results showed thatm for both domains, our method could provide higher accuracy and a higher F1 score than complex, but non-transparent, deep-learning models, such as BERT and ERINE. [ABSTRACT FROM AUTHOR]

Accurate prediction of the DNA-binding protein (DBP) helps to uncover the underlying mechanism of DNA-protein interaction, thus contributing to the understanding of many biochemical processes. Although the computational methods have made great process, there is still much room for improving the DBP prediction. In this study, a lightweight and interpretable sequence-based method, named LBi-DBP, is well-designed to improve the performance of DBP prediction. In LBi-DBP, five sequence feature sources are first employed to extract important sequence context information using a slight bidirectional long short-term memory (BiLSTM)-based neural network module; then, a multi-layer perceptron (MLP) module is utilized to learn the optimal/sub-optimal parameters with sequence context information extracted by BiLSTM module and pseudo sequence order feature as inputs. Experimental results on two independent test sets, i.e., UniSwiss-Tst and PDB2272, demonstrate that LBi-DBP can achieve Matthew's correlation coefficient values of 0.762 and 0.574 respectively, which are both higher than those of the second-best existing DBP prediction method (0.741 and 0.424). Meanwhile, good prediction result on a disease-related test set also shows the value of LBi-DBP. Through interpretability analysis, we find that the BiLSTM module can focus on the residues in real native DNA-binding domain region without prior knowledge, thus improving DNA-binding protein prediction performance. [ABSTRACT FROM AUTHOR]

In the multi-organ segmentation task of medical images, there are some challenging issues such as the complex background, blurred boundaries between organs, and the larger scale difference in volume. Due to the local receptive fields of conventional convolution operations, it is difficult to obtain desirable results by directly using them for multi-organ segmentation. While Transformer-based models have global information, there is a significant dependency on hardware because of the high computational demands. Meanwhile, the depthwise convolution with large kernel can capture global information and have less computational requirements. Therefore, to leverage the large receptive field and reduce model complexity, we propose a novel CNN-based approach, namely adjacent-scale fusion U-Net with large kernel (ASF-LKUNet) for multi-organ segmentation. We utilize a u-shaped encoder-decoder as the base architecture of ASF-LKUNet. In the encoder path, we design the large kernel residual block, which combines the large and small kernels and can simultaneously capture the global and local features. Furthermore, for the first time, we propose an adjacent-scale fusion and large kernel GRN channel attention that incorporates the low-level details with the high-level semantics by the adjacent-scale feature and then adaptively focuses on the more global and meaningful channel information. Extensive experiments and interpretability analysis are made on the Synapse multi-organ dataset (Synapse) and the ACDC cardiac multi-structure dataset (ACDC). Our proposed ASF-LKUNet achieves 88.41% and 89.45% DSC scores on the Synapse and ACDC datasets, respectively, with 17.96M parameters and 29.14 GFLOPs. These results show that our method achieves superior performance with favorable lower complexity against ten competing approaches.ASF-LKUNet is superior to various competing methods and has less model complexity. Code and the trained models have been released on GitHub., Competing Interests: Declaration of competing interest The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper., (Copyright © 2024 Elsevier Ltd. All rights reserved.)

Canonical Correlation Analysis (CCA) has been widely applied to study correlations between neuroimaging data and behavioral data. Practical use of CCA typically requires dimensionality reduction with, for example, Principal Components Analysis (PCA), however, this can result in CCA components that are difficult to interpret. In this paper, we introduce a Domain-driven Dimension Reduction (DDR) method, reducing the dimensionality of the original datasets and combining human knowledge of the structure of the variables studied. We apply the method to the Human Connectome Project S1200 release and compare standard PCA across all variables with DDR applied to individual classes of variables, finding that DDR-CCA results are more stable and interpretable, allowing the contribution of each class of variable to be better understood. By carefully designing the analysis pipeline and cross-validating the results, we offer more insights into the interpretation of CCA applied to brain-behavior data. [ABSTRACT FROM AUTHOR]

• Contrastive learning improves the data utilization of belief representations in POMDPs. • The bottleneck belief in the bidirectional model effectively constrains the upper bound of prediction errors. • The bidirectional match regularization (BMR) is derived to stabilize the training process. • Interpretability analyses show that the learned representations can track beliefs accurately. The key to solving the complex, partially observable Markov decision process (POMDP) with high-dimensional observations lies in explicit belief representations. However, the existing methods generally adopted the black-box model for shaping beliefs, which is inefficient and lacks interpretability. Due to this reason, the action-dependent bidirectional contrastive predictive coding (BCPC|Action) is proposed in this paper, in which the observation features are extracted efficiently through self-supervised contrastive learning. Owing to the bottleneck belief constraints in the bidirectional model, the upper bound of prediction errors is effectively reduced. Besides, the forward prediction is optimized by the guidance of an easier trainable backward prediction; thus, the bidirectional match regularization (BMR) could be derived for stabilizing the training process. More importantly, the interpretability of the learned belief representation is thoroughly explored based on the gradient truncation. Simulation results verify the effectiveness of the presented method; apart from achieving highly accurate belief tracking, the state uncertainties could be characterized reasonably, which provides a guarantee for solving the POMDP optimal policy for downstream tasks. [ABSTRACT FROM AUTHOR]

Accurate wind speed forecasting holds the potential to optimize wind farm design and enhance the utilization of wind energy resources. However, the majority of current research has encountered three issues: (1) Feature selection is independent of the sub-models and does not consider the sensitivity of the model to features, resulting in all sub-models using the same combination of features for training and prediction. (2) Sub-models are chosen only on the basis of their fitting ability, neglecting the distinctions between them. (3) There is a lack of interpretability analysis regarding the model's forecasting. It is particularly important to make interpretable analyses for predictive models, as it not only helps to increase the trustworthiness of the model, but also assist in fine-tuning the prediction model and identifying wind speed-related datasets. To address these issues, this paper proposes a novel method for combining wind speed forecasts. The approach comprises five steps: data noise reduction, feature selection, sub-model selection, sub-model combination, and interpretability analysis. (1) In the data noise reduction module, noise reduction is carried out on the raw data based on a combination of singular spectral analysis (SSA) and intrinsic computing expressive empirical mode decomposition with adaptive noise (ICEEMDAN). (2) In the feature selection module, a quantitative feature importance calculation approach is proposed to select the optimal feature input for eight different sub-models, back propagation neural network (BPNN), extreme learning machine (ELM), Elman, wavelet neural network (WNN), generalized regression neural network (GRNN), long short-term memory (LSTM), gate recurrent unit (GRU) and recurrent neural network (RNN). The method calculates the average marginal effect of each feature output in each sub-model and its fluctuation as the feature importance. (3) In the sub-model selection module, we propose an optimal comprehensive sub-model selection metric (CSMS), it combines the fitting ability and the variability of the sub-model. (4) In the combination module, a multi-objective coati optimization algorithm (MOCOA) is used to combine the selected optimal sub-models. (5) In the interpretability analysis, we leverage the deletion diagnosis method to explain and analyze the forecasting behavior of sub-models from both global and local sample perspectives. The experimental results show that compared with the other models, the proposed model is optimal, the MAPE is 8.4571 %, the R 2 can reach 0.9755, and it can be concluded at a pr = 99 % confidence level that the proposed model and the sub-models have a significant difference in performance. [ABSTRACT FROM AUTHOR]

Canonical Correlation Analysis (CCA) has been widely applied to study correlations between neuroimaging data and behavioral data. Practical use of CCA typically requires dimensionality reduction with, for example, Principal Components Analysis (PCA), however, this can result in CCA components that are difficult to interpret. In this paper, we introduce a Domain-driven Dimension Reduction (DDR) method, reducing the dimensionality of the original datasets and combining human knowledge of the structure of the variables studied. We apply the method to the Human Connectome Project S1200 release and compare standard PCA across all variables with DDR applied to individual classes of variables, finding that DDR-CCA results are more stable and interpretable, allowing the contribution of each class of variable to be better understood. By carefully designing the analysis pipeline and cross-validating the results, we offer more insights into the interpretation of CCA applied to brain-behavior data.

Current failure prediction methods of bearings have less uncertainty analysis with interpretability, less correlation analysis between degradation characteristics and prediction error. Moreover, there are multiple degradation stages in entire life cycle and prediction performance cannot meet practical demands. Therefore, this paper proposes a new approach for failure prediction of bearings. The change point detection method achieves multi-stage division of degradation data. The improved hybrid deep forest with best dissimilarity sequence (BDS) is studied and a new pretrained algorithm with pruning operation is developed. The convergence theorem is proved. A novel multi-stage failure prediction algorithm based on improved hybrid deep forest, hypothesis testing and interpretability analysis, is developed to get better prediction result. The time complexity of proposed algorithm is analyzed. The datasets of NASA and FEMTO-ST institute are utilized and experimental results show that: 1) Our approach with model interpretability has better prediction performances than support vector machine (SVR), recurrent neural network (RNN), long short-term memory (LSTM), and deep forest (DF); 2) The non-normal distribution characteristics, monotonic degradation trend and effect size of multiple stages are analyzed based on hypothesis testing methods; 3) The positive and inverse relation analysis achieves the correlation interpretability between multi-stage degradation characteristics and failure prediction results. • EWMA, CUSUM, and K-means clustering are used to obtain real change points for multi-stage division. • An improved hybrid deep forest model with BDS is presented to improve prediction performance. • This paper proposes a new pretrained algorithm to achieve a better tradeoff between accuracy and runtime cost. • A multi-stage failure prediction algorithm with model interpretability is developed. • Hypothesis testing and correlation analysis are utilized to enhance the interpretability of degradation characteristics. [ABSTRACT FROM AUTHOR]

For the diagnosis and outcome prediction of gastric cancer (GC), machine learning methods based on whole slide pathological images (WSIs) have shown promising performance and reduced the cost of manual analysis. Nevertheless, accurate prediction of GC outcome may rely on multiple modalities with complementary information, particularly gene expression data. Thus, there is a need to develop multimodal learning methods to enhance prediction performance. In this paper, we collect a dataset from Ruijin Hospital and propose a multimodal learning method for GC diagnosis and outcome prediction, called GaCaMML, which is featured by a cross-modal attention mechanism and Per-Slide training scheme. Additionally, we perform feature attribution analysis via integrated gradient (IG) to identify important input features. The proposed method improves prediction accuracy over the single-modal learning method on three tasks, i.e., survival prediction (by 4.9% on C-index), pathological stage classification (by 11.6% on accuracy), and lymph node classification (by 12.0% on accuracy). Especially, the Per-Slide strategy addresses the issue of a high WSI-to-patient ratio and leads to much better results compared with the Per-Person training scheme. For the interpretable analysis, we find that although WSIs dominate the prediction for most samples, there is still a substantial portion of samples whose prediction highly relies on gene expression information. This study demonstrates the great potential of multimodal learning in GC-related prediction tasks and investigates the contribution of WSIs and gene expression, respectively, which not only shows how the model makes a decision but also provides insights into the association between macroscopic pathological phenotypes and microscopic molecular features. [Display omitted] • GaCaMML outperforms the single-modal learning method across three tasks on accuracy. • Per-Slide strategy enhances prediction results by optimizing WSI-to-patient ratio. • WSI features prevail in gastric cancer outcome prediction for most samples. • Certain genes like MMPs and TP53 significantly impact prediction results. [ABSTRACT FROM AUTHOR]

Despite achieving success in many domains, deep learning models remain mostly black boxes, especially in electroencephalogram (EEG)-related tasks. Meanwhile, understanding the reasons behind model predictions is quite crucial in assessing trust and performance promotion in EEG-related tasks. In this work, we explore the use of representative interpretable models to analyze the learning behavior of convolutional neural networks (CNN) in EEG-based emotion recognition. According to the interpretable analysis, we find that similar features captured by our model and state-of-the-art model are consistent with previous brain science findings. Next, we propose a new model by integrating brain science knowledge with the interpretability analysis results in the learning process. Our knowledge-integrated model achieves better recognition accuracy on standard EEG-based recognition datasets.

Coronavirus Disease 2019 (COVID-19) continues to spread throughout the world, and it is necessary for us to implement effective methods to prevent and control the spread of the epidemic. In this paper, we propose a new model called Spatial–Temporal Attention Graph Convolutional Networks (STAGCN) that can analyze the long-term trend of the COVID-19 epidemic with high accuracy. The STAGCN employs a spatial graph attention network layer and a temporal gated attention convolutional network layer to capture the spatial and temporal features of infectious disease data, respectively. While the new model inherits the symmetric "space-time space" structure of Spatial–Temporal Graph Convolutional Networks (STGCN), it enhances its ability to identify infectious diseases using spatial–temporal correlation features by replacing the graph convolutional network layer with a graph attention network layer that can pay more attention to important features based on adaptively adjusted feature weights at different time points. The experimental results show that our model has the lowest error rate compared to other models. The paper also analyzes the prediction results of the model using interpretable analysis methods to provide a more reliable guide for the decision-making process during epidemic prevention and control. [ABSTRACT FROM AUTHOR]

Machine learning models have been widely used for studying thermal sensations. However, the black‐box properties of machine learning models lead to the lack of model transparency, and existing explanations for the thermal sensation models are generally flawed in terms of the perspectives of interpretable methods. In this study, we perform an interpretability analysis using the "SHapley Additive exPlanation" (SHAP) from game theory for thermal sensation machine learning models. The effects of different features on thermal sensations and typical decision routes in the models are investigated from both local and global perspectives, and the properties of correlation between features and thermal sensations and decision routes within machine learning models are summarized. The differences in the effects of features across samples reflect the effects of features on thermal sensations not only can be demonstrated by significant magnitudes but also by differentiation. The effects of features on thermal sensations often appear in the form of combinations of two to four features, which determine the final thermal sensation in most cases. Therefore, the neutral environment may actually be a dynamic high‐dimensional space consisting of certain combinations of features in certain ranges with changing shapes. [ABSTRACT FROM AUTHOR]

We develop a Bayesian network (LASSO‐BN) model for firm bankruptcy prediction. We select financial ratios via the Least Absolute Shrinkage Selection Operator (LASSO), establish the BN topology, and estimate model parameters. Our empirical results, based on 32,344 US firms from 1961–2018, show that the LASSO‐BN model outperforms most alternative methods except the deep neural network. Crucially, the model provides a clear interpretation of its internal functionality by describing the logic of how conditional default probabilities are obtained from selected variables. Thus our model represents a major step towards interpretable machine learning models with strong performance and is relevant to investors and policymakers. [ABSTRACT FROM AUTHOR]

In Natural Language Processing (NLP), deep-learning neural networks have superior performance but pose transparency and explainability barriers, due to their black box nature, and, thus, there is lack of trustworthiness. On the other hand, classical machine learning techniques are intuitive and easy to understand but often cannot perform satisfactorily. Fortunately, many research studies have recently indicated that the newly introduced model, Tsetlin Machine (TM), has reliable performance and, at the same time, enjoys human-level interpretability by nature, which is a promising approach to trade off effectiveness and interpretability. However, nearly all of the related works so far have concentrated on the English language, while research on other languages is relatively scarce. So, we propose a novel method, based on the TM model, in which the learning process is transparent and easily-understandable for Chinese NLP tasks. Our model can learn semantic information in the Chinese language by clauses. For evaluation, we conducted experiments in two domains, namely sentiment analysis and spam review detection. The experimental results showed thatm for both domains, our method could provide higher accuracy and a higher F1 score than complex, but non-transparent, deep-learning models, such as BERT and ERINE.

Although research on applying machine learning to the performance of the built environment has been advancing considerably, outdoor environment prediction models still need to be more accurate. In this study, I investigated hybrid-driven methods for developing environmental performance prediction models and studied how machine learning algorithms may interpret spatial information in the context of an environmental performance simulation challenge. The simulation of the Universal Thermal Climate Index (UTCI) for outdoor applications served as an example. Specifically, I designed two different network structures, each with six neural network models. These neural network models were built with various numbers of layers, convolutional kernel sizes, and convolutional kernel layers. As shown by these models’ training results, I investigated the effect of model parameter settings on performance. In addition, I conducted interpretable analysis through the visual observation of hidden internal layers. The use of multilayer and small convolutional kernels, as well as an increase in the amount of training data, may be the reason neural network prediction performance was improved. From the perspective of interpretability analysis, the convolutional layer can more accurately analyze building space problems, and full connection layers focus more on the regression between the spatial features and performance results. This “space analysis → data regression” network structure can be expanded to wind environment forecasting or heat environment in the future.

The peptide therapeutics market is providing new opportunities for the biotechnology and pharmaceutical industries. Therefore, identifying therapeutic peptides and exploring their properties are important. Although several studies have proposed different machine learning methods to predict peptides as being therapeutic peptides, most do not explain the decision factors of model in detail. In this work, an Interpretable Therapeutic Peptide Prediction (ITP-Pred) model based on efficient feature fusion was developed. First, we proposed three kinds of feature descriptors based on sequence and physicochemical property encoded, namely amino acid composition (AAC), group AAC and coding autocorrelation, and concatenated them to obtain the feature representation of therapeutic peptide. Then, we input it into the CNN-Bi-directional Long Short-Term Memory (BiLSTM) model to automatically learn recognition of therapeutic peptides. The cross-validation and independent verification experiments results indicated that ITP-Pred has a higher prediction performance on the benchmark dataset than other comparison methods. Finally, we analyzed the output of the model from two aspects: sequence order and physical and chemical properties, mining important features as guidance for the design of better models that can complement existing methods. [ABSTRACT FROM AUTHOR]

A supervised learning model combined with genetic algorithm is proposed to predict charge transfer efficiency of nanodiamonds. The model is chosen among ten models whose parameters are modified by genetic algorithm with ten-fold cross-validation, ensuring the accuracy of model. Generalization ability and reliability are further verified by prediction error and consistency tests of twin nanodiamonds. A hydrogenated surface nanodiamond with low ionization potential and a clean nanodiamond with low electron affinity are designed based on particle swarm optimization. It is further found via SHAP analysis that electron affinity is inhibited by surfaces with a {111} surface below 30 % or a {100} surface above 80 %. Similarly, ionization potential is reduced when the hybrid ratios of sp1 and sp2 are separately greater than 0.501 % or lower than 5.690 %. [Display omitted] [ABSTRACT FROM AUTHOR]

Continuous effluent quality prediction in wastewater treatment processes is crucial to proactively reduce the risks to the environment and human health. However, wastewater treatment is an extremely complex process controlled by several uncertain, interdependent, and sometimes poorly characterized physico-chemical-biological process parameters. In addition, there are substantial spatiotemporal variations, uncertainties, and high non-linear interactions among the water quality parameters and process variables involved in the treatment process. Such complexities hinder efficient monitoring, operation, and management of wastewater treatment plants under normal and abnormal conditions. Typical mathematical and statistical tools most often fail to capture such complex interrelationships, and therefore data-driven techniques offer an attractive solution to effectively quantify the performance of wastewater treatment plants. Although several previous studies focused on applying regression-based data-driven models (e.g., artificial neural network) to predict some wastewater treatment effluent parameters, most of these studies employed a limited number of input variables to predict only one or two parameters characterizing the effluent quality (e.g., chemical oxygen demand (COD) and/or suspended solids (SS)). Harnessing the power of Artificial Intelligence (AI), the current study proposes multi-gene genetic programming (MGGP)-based models, using a dataset obtained from an operational wastewater treatment plant, deploying membrane aerated biofilm reactor, to predict the filtrated COD, ammonia (NH 4), and SS concentrations along with the carbon-to-nitrogen ratio (C/N) within the effluent. Input features included a set of process variables characterizing the influent quality (e.g., filtered COD, NH 4 , and SS concentrations), water physics and chemistry parameters (e.g., temperature and pH), and operation conditions (e.g., applied air pressure). The developed MGGP-based models accurately reproduced the observations of the four output variables with correlation coefficient values that ranged between 0.98 and 0.99 during training and between 0.96 and 0.99 during testing, reflecting the power of the developed models in predicting the quality of the effluent from the treatment system. Interpretability analyses were subsequently deployed to confirm the intuitive understanding of input-output interrelations and to identify the governing parameters of the treatment process. The developed MGGP-based models can facilitate the AI-driven monitoring and management of wastewater treatment plants through devising optimal rapid operation and control schemes and assisting the plants' operators in maintaining proper performance of the plants under various normal and disruptive operational conditions. [Display omitted] • An interdependence analysis was done on a dataset from wastewater treatment system. • MGGP-based models were developed to accurately predict the effluent concentrations. • Interpretability study was done to confirm physical connections between variables. • MGGP models are an efficient alternative for operating wastewater treatment plants. [ABSTRACT FROM AUTHOR]

Coronavirus Disease 2019 (COVID-19) continues to spread throughout the world, and it is necessary for us to implement effective methods to prevent and control the spread of the epidemic. In this paper, we propose a new model called Spatial–Temporal Attention Graph Convolutional Networks (STAGCN) that can analyze the long-term trend of the COVID-19 epidemic with high accuracy. The STAGCN employs a spatial graph attention network layer and a temporal gated attention convolutional network layer to capture the spatial and temporal features of infectious disease data, respectively. While the new model inherits the symmetric “space-time space” structure of Spatial–Temporal Graph Convolutional Networks (STGCN), it enhances its ability to identify infectious diseases using spatial–temporal correlation features by replacing the graph convolutional network layer with a graph attention network layer that can pay more attention to important features based on adaptively adjusted feature weights at different time points. The experimental results show that our model has the lowest error rate compared to other models. The paper also analyzes the prediction results of the model using interpretable analysis methods to provide a more reliable guide for the decision-making process during epidemic prevention and control.