Search

Objective: Research into the effectiveness and applicability of deep learning, radiomics, and their integrated models based on Magnetic Resonance Imaging (MRI) for preoperative differentiation between Primary Central Nervous System Lymphoma (PCNSL) and Glioblastoma (GBM), along with an exploration of the interpretability of these models. Materials and methods: A retrospective analysis was performed on MRI images and clinical data from 261 patients across two medical centers. The data were split into a training set (n = 153, medical center 1) and an external test set (n = 108, medical center 2). Radiomic features were extracted using Pyradiomics to build the Radiomics Model. Deep learning networks, including the transformer-based MobileVIT Model and Convolutional Neural Networks (CNN) based ConvNeXt Model, were trained separately. By applying the "late fusion" theory, the radiomics model and deep learning model were fused to produce the optimal Max-Fusion Model. Additionally, Shapley Additive exPlanations (SHAP) and Grad-CAM were employed for interpretability analysis. Results: In the external test set, the Radiomics Model achieved an Area under the receiver operating characteristic curve (AUC) of 0.86, the MobileVIT Model had an AUC of 0.91, the ConvNeXt Model demonstrated an AUC of 0.89, and the Max-Fusion Model showed an AUC of 0.92. The Delong test revealed a significant difference in AUC between the Max-Fusion Model and the Radiomics Model (P = 0.02). Conclusion: The Max-Fusion Model, combining different models, presents superior performance in distinguishing PCNSL and GBM, highlighting the effectiveness of model fusion for enhanced decision-making in medical applications. Clinical Relevance Statement: The preoperative non-invasive differentiation between PCNSL and GBM assists clinicians in selecting appropriate treatment regimens and clinical management strategies. [ABSTRACT FROM AUTHOR]

Purpose: Vaccine supply planning remains a critical and challenging issue for developing countries due to the limited resources and fluctuating vaccine demand, reflected by the sudden increase and decrease in the number of vaccinated people. Therefore, it is essential to accurately predict vaccination rates and interpret the prediction results to anticipate vaccine demand. Previous research focused on the prediction of vaccination rates by considering vaccination data and overlooked the importance of including features that highly affect the prediction results. Methods: This research proposes an interpretable Convolutional Neural Network and Long Short-term Memory (CNN-LSTM) model for COVID-19 vaccination rate prediction, specifically for developing countries. The proposed model includes a three-stage feature engineering process that utilizes domain knowledge to construct new features of disease spread and the strictness of government policies. The temporal correlation of the features is captured to improve the prediction accuracy. Results: The case studies were conducted on Nigeria, Nepal, and Peru's datasets to ensure the model's effectiveness and generalizability. The results indicate the effectiveness of the model, showing an average 30% decrease in the Mean Absolute Percentage Error (MAPE) with the enhanced feature engineering process and 16% additional improvement after combining the CNN with LSTM compared to the standalone LSTM model. Conclusion: The interpretability of the CNN-LSTM model is a key aspect considered in this research to explain the prediction results. The model's interpretability aims to identify important features affecting the prediction results and provide government and vaccine companies with guidelines in the decision-making for vaccine supply planning in developing countries. [ABSTRACT FROM AUTHOR]

Bladder lesion commonly occurs in patients with benign prostatic hyperplasia (BPH), and the routine screening of bladder lesion is vital for its timely detection and treatment, in which the risk of bladder lesion progression can be effectively alleviated. However, current clinical methods are inconvenient for routine screening. In this study, we proposed a convenient routine screening method to diagnose bladder lesions based on several clinical risk factors, which can be obtained through non-invasive, easy-to-operate, and low-cost examinations. The contribution of each clinical risk factor was further quantitatively analyzed to understand their impact on diagnostic decision-making. Based on a cohort study of 253 BPH patients with or without bladder lesions, the proposed diagnostic model achieved high accuracy using these clinical risk factors. Bladder compliance, maximum flow rate (Qmax), prostate specific antigen (PSA), and postvoid residual (PVR) were identified as the four most important clinical risk factors. To the best of our knowledge, this is the innovative research to predict bladder lesions based on the risk factors and quantitatively reveal their contributions to diagnostic decision-making. The proposed model has the potential to serve as an effective routine screening tool for bladder lesions in BPH patients, enabling early intervention to prevent lesion progression and improve the quality of life. [ABSTRACT FROM AUTHOR]

Sepsis is a leading cause of death among intensive care unit patients. Early sepsis prediction, which primarily relies on advanced artificial intelligence technology, is an important phase in sepsis management. In previous sepsis prediction research, tremendous time and effort are spent in building high accurate tree ensemble models. However, due to the "black box" nature of tree ensemble models, clinicians typically reluctant to use them in clinical practice. To fill the gap, a novel multi-objective rule extraction method is proposed to trade off the predictive performance and the interpretability of tree ensemble model for sepsis prediction. The proposed method encompasses two phases: rule extraction utilizing a novel multi-objective binary differential evolution algorithm (MOBDE), and rule simplification employing a greedy heuristic method. In addition, rule extraction incorporates a Tchebycheff-based competitive learning strategy and a binary mutation operation aimed at enhancing the search ability of MOBDE. Rule simplification leveraging the proposed greedy heuristic method, streamlines the extracted rules from the previous phase by eliminating redundant constraints. The 2019 PhysioNet/CinC Challenge dataset, a well-known dataset in sepsis prediction, is used to evaluate the effectiveness of the proposed method. The experimental results show that the proposed method is more interpretable and has higher accuracy than some of commonly used rule-based models for sepsis prediction. The proposed method can be popularized to practical sepsis treatment, and assist doctors in sepsis diagnosis with the comprehensible decision-making support. [ABSTRACT FROM AUTHOR]

Abstract Bladder lesion commonly occurs in patients with benign prostatic hyperplasia (BPH), and the routine screening of bladder lesion is vital for its timely detection and treatment, in which the risk of bladder lesion progression can be effectively alleviated. However, current clinical methods are inconvenient for routine screening. In this study, we proposed a convenient routine screening method to diagnose bladder lesions based on several clinical risk factors, which can be obtained through non-invasive, easy-to-operate, and low-cost examinations. The contribution of each clinical risk factor was further quantitatively analyzed to understand their impact on diagnostic decision-making. Based on a cohort study of 253 BPH patients with or without bladder lesions, the proposed diagnostic model achieved high accuracy using these clinical risk factors. Bladder compliance, maximum flow rate (Qmax), prostate specific antigen (PSA), and postvoid residual (PVR) were identified as the four most important clinical risk factors. To the best of our knowledge, this is the innovative research to predict bladder lesions based on the risk factors and quantitatively reveal their contributions to diagnostic decision-making. The proposed model has the potential to serve as an effective routine screening tool for bladder lesions in BPH patients, enabling early intervention to prevent lesion progression and improve the quality of life.

Abstract Background Pneumonia, a leading cause of morbidity and mortality worldwide, often necessitates Intensive Care Unit (ICU) admission. Accurate prediction of pneumonia mortality is crucial for tailored prevention and treatment plans. However, existing mortality prediction models face limited adoption in clinical practice due to their lack of interpretability. Objective This study aimed to develop an interpretable model for predicting pneumonia mortality in ICUs. Leveraging the Shapley Additive Explanation (SHAP) method, we sought to elucidate the Extreme Gradient Boosting (XGBoost) model and identify prognostic factors for pneumonia. Methods Conducted as a retrospective cohort study, we utilized electronic health records from the eICU-CRD (2014–2015) for all adult pneumonia patients. The first 24 h of each ICU admission records were considered, with 70% of the dataset allocated for model training and 30% for validation. The XGBoost model was employed, and performance was assessed using the area under the receiver operating characteristic curve (AUC). The SHAP method provided insights into the XGBoost model. Results Among 10,962 pneumonia patients, in-hospital mortality was 16.33%. The XGBoost model demonstrated superior predictive performance (AUC: 0.778 ± 0.016)) compared to traditional scoring systems and other machine learning method, which achieved an improvement of 10% points. SHAP analysis identified Aspartate Aminotransferase (AST) as the most crucial predictor. Conclusions Interpretable predictive models enhance mortality risk assessment for pneumonia patients in the ICU, fostering transparency. AST emerged as the foremost predictor, followed by patient age, albumin, BMI et al. These insights, rooted in strong correlations with mortality, facilitate improved clinical decision-making and resource allocation.

Abstract Background The advances in deep learning-based pathological image analysis have invoked tremendous insights into cancer prognostication. Still, lack of interpretability remains a significant barrier to clinical application. Methods We established an integrative prognostic neural network for intrahepatic cholangiocarcinoma (iCCA), towards a comprehensive evaluation of both architectural and fine-grained information from whole-slide images. Then, leveraging on multi-modal data, we conducted extensive interrogative approaches to the models, to extract and visualize the morphological features that most correlated with clinical outcome and underlying molecular alterations. Results The models were developed and optimized on 373 iCCA patients from our center and demonstrated consistent accuracy and robustness on both internal (n = 213) and external (n = 168) cohorts. The occlusion sensitivity map revealed that the distribution of tertiary lymphoid structures, the geometric traits of the invasive margin, the relative composition of tumor parenchyma and stroma, the extent of necrosis, the presence of the disseminated foci, and the tumor-adjacent micro-vessels were the determining architectural features that impacted on prognosis. Quantifiable morphological vector extracted by CellProfiler demonstrated that tumor nuclei from high-risk patients exhibited significant larger size, more distorted shape, with less prominent nuclear envelope and textural contrast. The multi-omics data (n = 187) further revealed key molecular alterations left morphological imprints that could be attended by the network, including glycolysis, hypoxia, apical junction, mTORC1 signaling, and immune infiltration. Conclusions We proposed an interpretable deep-learning framework to gain insights into the biological behavior of iCCA. Most of the significant morphological prognosticators perceived by the network are comprehensible to human minds. Graphical Abstract

Abstract Objectives To develop and validate a novel interpretable artificial intelligence (AI) model that integrates radiomic features, deep learning features, and imaging features at multiple semantic levels to predict the prognosis of intracerebral hemorrhage (ICH) patients at 6 months post-onset. Materials and methods Retrospectively enrolled 222 patients with ICH for Non-contrast Computed Tomography (NCCT) images and clinical data, who were divided into a training cohort (n = 186, medical center 1) and an external testing cohort (n = 36, medical center 2). Following image preprocessing, the entire hematoma region was segmented by two radiologists as the volume of interest (VOI). Pyradiomics algorithm library was utilized to extract 1762 radiomics features, while a deep convolutional neural network (EfficientnetV2-L) was employed to extract 1000 deep learning features. Additionally, radiologists evaluated imaging features. Based on the three different modalities of features mentioned above, the Random Forest (RF) model was trained, resulting in three models (Radiomics Model, Radiomics-Clinical Model, and DL-Radiomics-Clinical Model). The performance and clinical utility of the models were assessed using the Area Under the Receiver Operating Characteristic Curve (AUC), calibration curve, and Decision Curve Analysis (DCA), with AUC compared using the DeLong test. Furthermore, this study employs three methods, Shapley Additive Explanations (SHAP), Grad-CAM, and Guided Grad-CAM, to conduct a multidimensional interpretability analysis of model decisions. Results The Radiomics-Clinical Model and DL-Radiomics-Clinical Model exhibited relatively good predictive performance, with an AUC of 0.86 [95% Confidence Intervals (CI): 0.71, 0.95; P

Background: Pneumonia, a leading cause of morbidity and mortality worldwide, often necessitates Intensive Care Unit (ICU) admission. Accurate prediction of pneumonia mortality is crucial for tailored prevention and treatment plans. However, existing mortality prediction models face limited adoption in clinical practice due to their lack of interpretability. Objective: This study aimed to develop an interpretable model for predicting pneumonia mortality in ICUs. Leveraging the Shapley Additive Explanation (SHAP) method, we sought to elucidate the Extreme Gradient Boosting (XGBoost) model and identify prognostic factors for pneumonia. Methods: Conducted as a retrospective cohort study, we utilized electronic health records from the eICU-CRD (2014–2015) for all adult pneumonia patients. The first 24 h of each ICU admission records were considered, with 70% of the dataset allocated for model training and 30% for validation. The XGBoost model was employed, and performance was assessed using the area under the receiver operating characteristic curve (AUC). The SHAP method provided insights into the XGBoost model. Results: Among 10,962 pneumonia patients, in-hospital mortality was 16.33%. The XGBoost model demonstrated superior predictive performance (AUC: 0.778 ± 0.016)) compared to traditional scoring systems and other machine learning method, which achieved an improvement of 10% points. SHAP analysis identified Aspartate Aminotransferase (AST) as the most crucial predictor. Conclusions: Interpretable predictive models enhance mortality risk assessment for pneumonia patients in the ICU, fostering transparency. AST emerged as the foremost predictor, followed by patient age, albumin, BMI et al. These insights, rooted in strong correlations with mortality, facilitate improved clinical decision-making and resource allocation. [ABSTRACT FROM AUTHOR]

Straw burning is recognized as a significant contributor to deteriorating air quality, but its specific impacts, particularly on PM2.5 concentrations, are still not fully understood or quantified. In this study, we conducted a detailed examination of the spatial and temporal patterns of straw burning in Heilongjiang Province, China—a key agricultural area—utilizing high-resolution fire-point data from the Fengyun-3 satellite. We subsequently employed random forest (RF) models alongside Shapley Additive Explanations (SHAPs) to systematically evaluate the impact of various determinants, including straw burning (as indicated by crop fire-point data), meteorological conditions, and aerosol optical depth (AOD), on PM2.5 levels across spatial and temporal dimensions. Our findings indicated a statistically nonsignificant downward trend in the number of crop fires in Heilongjiang Province from 2015 to 2023, with hotspots mainly concentrated in the western and southern parts of the province. On a monthly scale, straw burning was primarily observed from February to April and October to November—which are critical periods in the agricultural calendar—accounting for 97% of the annual fire counts. The RF models achieved excellent performance in predicting PM2.5 levels, with R2 values of 0.997 for temporal and 0.746 for spatial predictions. The SHAP analysis revealed the number of fire points to be the key determinant of temporal PM2.5 variations during straw-burning periods, explaining 72% of the variance. However, the significance was markedly reduced in the spatial analysis. This study leveraged machine learning and interpretable modeling techniques to provide a comprehensive understanding of the influence of straw burning on PM2.5 levels, both temporally and spatially. The detailed analysis offers valuable insights for policymakers to formulate more targeted and effective strategies to combat air pollution. [ABSTRACT FROM AUTHOR]

We present a machine learning framework capable of consistently inferring mathematical expressions of hyperelastic energy functionals for incompressible materials from sparse experimental data and physical laws. To achieve this goal, we propose a polyconvex neural additive model (PNAM) that enables us to express the hyperelastic model in a learnable feature space while enforcing polyconvexity. An upshot of this feature space obtained via the PNAM is that (1) it is spanned by a set of univariate basis functions that can be re‐parametrized with a more complex mathematical form, and (2) the resultant elasticity model is guaranteed to fulfill the polyconvexity, which ensures that the acoustic tensor remains elliptic for any deformation. To further improve the interpretability, we use genetic programming to convert each univariate basis into a compact mathematical expression. The resultant multi‐variable mathematical models obtained from this proposed framework are not only more interpretable but are also proven to fulfill physical laws. By controlling the compactness of the learned symbolic form, the machine learning‐generated mathematical model also requires fewer arithmetic operations than its deep neural network counterparts during deployment. This latter attribute is crucial for scaling large‐scale simulations where the constitutive responses of every integration point must be updated within each incremental time step. We compare our proposed model discovery framework against other state‐of‐the‐art alternatives to assess the robustness and efficiency of the training algorithms and examine the trade‐off between interpretability, accuracy, and precision of the learned symbolic hyperelastic models obtained from different approaches. Our numerical results suggest that our approach extrapolates well outside the training data regime due to the precise incorporation of physics‐based knowledge. [ABSTRACT FROM AUTHOR]

Copyright of Computer Measurement & Control is the property of Magazine Agency of Computer Measurement & Control and its content may not be copied or emailed to multiple sites or posted to a listserv without the copyright holder's express written permission. However, users may print, download, or email articles for individual use. This abstract may be abridged. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material for the full abstract. (Copyright applies to all Abstracts.)

Background: The advances in deep learning-based pathological image analysis have invoked tremendous insights into cancer prognostication. Still, lack of interpretability remains a significant barrier to clinical application. Methods: We established an integrative prognostic neural network for intrahepatic cholangiocarcinoma (iCCA), towards a comprehensive evaluation of both architectural and fine-grained information from whole-slide images. Then, leveraging on multi-modal data, we conducted extensive interrogative approaches to the models, to extract and visualize the morphological features that most correlated with clinical outcome and underlying molecular alterations. Results: The models were developed and optimized on 373 iCCA patients from our center and demonstrated consistent accuracy and robustness on both internal (n = 213) and external (n = 168) cohorts. The occlusion sensitivity map revealed that the distribution of tertiary lymphoid structures, the geometric traits of the invasive margin, the relative composition of tumor parenchyma and stroma, the extent of necrosis, the presence of the disseminated foci, and the tumor-adjacent micro-vessels were the determining architectural features that impacted on prognosis. Quantifiable morphological vector extracted by CellProfiler demonstrated that tumor nuclei from high-risk patients exhibited significant larger size, more distorted shape, with less prominent nuclear envelope and textural contrast. The multi-omics data (n = 187) further revealed key molecular alterations left morphological imprints that could be attended by the network, including glycolysis, hypoxia, apical junction, mTORC1 signaling, and immune infiltration. Conclusions: We proposed an interpretable deep-learning framework to gain insights into the biological behavior of iCCA. Most of the significant morphological prognosticators perceived by the network are comprehensible to human minds. [ABSTRACT FROM AUTHOR]

Objectives: To develop and validate a novel interpretable artificial intelligence (AI) model that integrates radiomic features, deep learning features, and imaging features at multiple semantic levels to predict the prognosis of intracerebral hemorrhage (ICH) patients at 6 months post-onset. Materials and methods: Retrospectively enrolled 222 patients with ICH for Non-contrast Computed Tomography (NCCT) images and clinical data, who were divided into a training cohort (n = 186, medical center 1) and an external testing cohort (n = 36, medical center 2). Following image preprocessing, the entire hematoma region was segmented by two radiologists as the volume of interest (VOI). Pyradiomics algorithm library was utilized to extract 1762 radiomics features, while a deep convolutional neural network (EfficientnetV2-L) was employed to extract 1000 deep learning features. Additionally, radiologists evaluated imaging features. Based on the three different modalities of features mentioned above, the Random Forest (RF) model was trained, resulting in three models (Radiomics Model, Radiomics-Clinical Model, and DL-Radiomics-Clinical Model). The performance and clinical utility of the models were assessed using the Area Under the Receiver Operating Characteristic Curve (AUC), calibration curve, and Decision Curve Analysis (DCA), with AUC compared using the DeLong test. Furthermore, this study employs three methods, Shapley Additive Explanations (SHAP), Grad-CAM, and Guided Grad-CAM, to conduct a multidimensional interpretability analysis of model decisions. Results: The Radiomics-Clinical Model and DL-Radiomics-Clinical Model exhibited relatively good predictive performance, with an AUC of 0.86 [95% Confidence Intervals (CI): 0.71, 0.95; P < 0.01] and 0.89 (95% CI: 0.74, 0.97; P < 0.01), respectively, in the external testing cohort. Conclusion: The multimodal explainable AI model proposed in this study can accurately predict the prognosis of ICH. Interpretability methods such as SHAP, Grad-CAM, and Guided Grad-Cam partially address the interpretability limitations of AI models. Integrating multimodal imaging features can effectively improve the performance of the model. Clinical relevance statement: Predicting the prognosis of patients with ICH is a key objective in emergency care. Accurate and efficient prognostic tools can effectively prevent, manage, and monitor adverse events in ICH patients, maximizing treatment outcomes. Key points: •A multimodal AI model combining imaging features, deep learning features, and radiomics features at three different semantic levels outperforms traditional radiomics models; •This study proposes an interpretable deep learning radiomics model that enables the model to maintain a certain level of interpretability while meeting high-performance requirements; [ABSTRACT FROM AUTHOR]

Abstract Timely, accurate, and reliable information is essential for decision-makers, emergency managers, and infrastructure operators during flood events. This study demonstrates that a proposed machine learning model, MaxFloodCast, trained on physics-based hydrodynamic simulations in Harris County, offers efficient and interpretable flood inundation depth predictions. Achieving an average $$R^2$$ R 2 of 0.949 and a Root Mean Square Error of 0.61 ft (0.19 m) on unseen data, it proves reliable in forecasting peak flood inundation depths. Validated against Hurricane Harvey and Tropical Storm Imelda, MaxFloodCast shows the potential in supporting near-time floodplain management and emergency operations. The model’s interpretability aids decision-makers in offering critical information to inform flood mitigation strategies, to prioritize areas with critical facilities and to examine how rainfall in other watersheds influences flood exposure in one area. The MaxFloodCast model enables accurate and interpretable inundation depth predictions while significantly reducing computational time, thereby supporting emergency response efforts and flood risk management more effectively.

The Kano model is an extensively used technique for understanding different types of customer preferences. It classifies product features based on the effects of their performance on the overall customer satisfaction. Compared to surveys, numerous online reviews can be easily collected at a lower cost. This paper proposes an explainable neural network-based approach for the Kano categorisation of product features from online reviews. First, product feature words are identified by clustering nouns based on word embedding. Subsequently, the sentiments of the product feature words are determined by conducting the Vader sentiment analysis. Finally, the effects of the sentiments of each product feature on the star rating are estimated using explainable neural networks. Based on their effects, the product features are classified into the Kano categories. A case study of three Fitbit models is performed to validate the proposed approach. The Kano categorisation by the proposed approach is compared with the results of a previous product feature word clustering and ensemble neural network-based method. The results exhibit that the former presents a more reliable performance than the latter. The proposed approach is automated after providing several hyperparameters and can assist companies in conducting the Kano analysis with increased speed and efficiency. [ABSTRACT FROM AUTHOR]

Timely, accurate, and reliable information is essential for decision-makers, emergency managers, and infrastructure operators during flood events. This study demonstrates that a proposed machine learning model, MaxFloodCast, trained on physics-based hydrodynamic simulations in Harris County, offers efficient and interpretable flood inundation depth predictions. Achieving an average R 2 of 0.949 and a Root Mean Square Error of 0.61 ft (0.19 m) on unseen data, it proves reliable in forecasting peak flood inundation depths. Validated against Hurricane Harvey and Tropical Storm Imelda, MaxFloodCast shows the potential in supporting near-time floodplain management and emergency operations. The model's interpretability aids decision-makers in offering critical information to inform flood mitigation strategies, to prioritize areas with critical facilities and to examine how rainfall in other watersheds influences flood exposure in one area. The MaxFloodCast model enables accurate and interpretable inundation depth predictions while significantly reducing computational time, thereby supporting emergency response efforts and flood risk management more effectively. [ABSTRACT FROM AUTHOR]

With the widespread application of machine learning models, researchers have gradually recognized the limitations of such methods. Most of these models are black-box models, resulting in poor interpretability. To address this issue, this paper proposed a rule-based interpretable model and rule reduction method based on ensemble learning models, which included generating optimized random forest models, discovering and reducing redundant rules, and other steps. Firstly, this paper proposed an evaluation method for random forest models, and optimized the key parameters of random forest models based on the idea of reinforcement learning, resulting in a more interpretable random forest model. Secondly, the rule sets extracted from the random forest model were subjected to redundancy elimination, resulting in a more concise rule set. Experimental results on public datasets show that the generated rule sets perform well in terms of prediction accuracy and interpretability. [ABSTRACT FROM AUTHOR]

ObjectivesTo investigate the value of interpretable machine learning model and nomogram based on clinical factors, MRI imaging features, and radiomic features to predict Ki-67 expression in primary central nervous system lymphomas (PCNSL).Materials and methodsMRI images and clinical information of 92 PCNSL patients were retrospectively collected, which were divided into 53 cases in the training set and 39 cases in the external validation set according to different medical centers. A 3D brain tumor segmentation model was trained based on nnU-NetV2, and two prediction models, interpretable Random Forest (RF) incorporating the SHapley Additive exPlanations (SHAP) method and nomogram based on multivariate logistic regression, were proposed for the task of Ki-67 expression status prediction.ResultsThe mean dice Similarity Coefficient (DSC) score of the 3D segmentation model on the validation set was 0.85. On the Ki-67 expression prediction task, the AUC of the interpretable RF model on the validation set was 0.84 (95% CI:0.81, 0.86; p

Energy demand prediction can benefit electricity consumers, distribution network service providers, and system operators. It heavily depends on auxiliary factors, such as weather parameters (e.g., ambient temperature), which makes this problem more complex. Moreover, for many industrial applications, instead of aiming for the highest prediction accuracy, a more easily understandable and interpretable model that can lead to higher accuracy against the baseline model is the priority. Therefore, this problem still requires more investigation, especially when there is a specified prediction baseline to be compared with. This paper proposes a machine learning (ML) based prediction framework that investigates how temperature combined with energy consumption and simple and interpretable ML methods can be used to provide more precise demand forecasts and thus baselines closer to actual load profiles. The proposed framework is tested on two different real-world energy demand datasets. The analysis shows that using a simple ML model, such as a polynomial regression model, results in a more accurate prediction than the current baselines used in the energy market. The proposed ML models are not black-box type models, and thus are easier to explain and interpret. The ML-based forecasted demand is used as a baseline for demand response (DR) and is compared with the existing baselines used in the demand response market of Australia’s national grid.

The evaluation and prediction of credit risk have always been a research hotspot to ensure the healthy and orderly development of the credit market. Most researchers use deep learning to predict credit risk. However, when training data are too small, deep learning models often lead to overfitting. Although we have a large amount of available training data, we often cannot ensure that the data are evenly distributed, which is still not conducive to model training. In addition, deep learning is often difficult to explain, and the unexplained model is often difficult to gain the trust of users, thus reducing the usefulness of the model. To solve these problems, we propose an integrated cross-domain credit default prediction network, called Transfer Light Gradient Boosting Machine (TrLightGBM), based on interpretable integration transfer. This network considers the weight of data from different domains in training and implements cross-domain credit default prediction by adjusting the weight. The experiment shows that our method TrLightGBM not only achieves the interpretability of the model to a certain extent but also has good performance.

Sepsis is among the leading causes of morbidity, mortality and high costs in the ICU. The early prediction and intervention of sepsis is a challenging task under strict time and cost constraints. In this paper, a novel High-order Markov Dynamic Bayesian Network (HMDBN) classifier with discrete features is presented for early prediction of sepsis at a high-order time point. The model structure is learned from the unrolled DBN by performing the K2 algorithm, and the features 'disappeared' in the prediction are eliminated using the VE method. Based on a few vital signs and laboratory results, an intuitive causal graph and indicating system are constructed to realize continuous prediction and probabilistic interpretation in real-time. Compared with other ten classical machine learning classifiers on evaluation metrics, HMDBN models have the highest AUROC scores on both internal tests and external validations for sepsis early prediction, and provide identifiable and interpretable results that allowing clinicians to immediately understand the reason for the prediction. [ABSTRACT FROM AUTHOR]

This study investigates the Shapley additive explanation (SHAP) of the extreme boosting (XGBoost) model for breast cancer diagnosis. The study employed Wisconsin’s breast cancer dataset, characterized by 30 features extracted from an image of a breast cell. SHAP module generated different explainer values representing the impact of a breast cancer feature on breast cancer diagnosis. The experiment computed SHAP values of 569 samples of the breast cancer dataset. The SHAP explanation indicates perimeter and concave points have the highest impact on breast cancer diagnosis. SHAP explains the XGB model diagnosis outcome showing the features affecting the XGBoost model. The developed XGB model achieves an accuracy of 98.42% [ABSTRACT FROM AUTHOR]

Soil temperature (ST) is a crucial parameter in Earth system science. Accurate ST predictions provide invaluable insights; however, the “black box” nature of many deep learning approaches limits their interpretability. In this study, we present the Encoder-Decoder Model with Interpretable Spatio-Temporal Component (ISDNM) to enhance both ST prediction accuracy and its spatio-temporal interpretability. The ISDNM combines a CNN-encoder-decoder and an LSTM-encoder-decoder to improve spatio-temporal feature representation. It further uses linear regression and Uniform Manifold Approximation and Projection (UMAP) techniques for clearer spatio-temporal visualization of ST. The results show that the ISDNM model had the highest R2 ranging from 0.886 to 0.963 and the lowest RMSE ranging from 6.086 m3/m3 to 12.533 m3/m3 for different climate regions, and demonstrated superior performance than all the other DL models like CNN, LSTM, ConvLSTM models. The predictable component highlighted the remarkable similarity between Medium fine and Very fine soils in China. Additional, May and November emerged as crucial months, acting as inflection points in the annual ST cycle, shaping ISDNM model’s prediction capabilities.

Significant energy savings can be realised from buildings if deviations from the usual operating conditions are detected early, and appropriate measures are taken. Building anomaly detection techniques automate identifying such instances by leveraging the high dimensional data collected from the installed smart meters. Autoencoders allow for dimensionality reduction and also model the underlying data distribution. However, these models treat features as independent quantities. In contrast, the current work investigates an attention mechanism with an autoencoder to include the correlations among the features. The value addition from the attention mechanism is demonstrated by comparing the model’s reconstruction ability with an ANN-based autoencoder on synthetic datasets. The study identifies that adding an attention layer enables the encoder–decoder architecture to be robust to outliers in training data, thereby reducing the preprocessing required. Further, the model is tested on a real-world dataset, and the attention maps generated from the model are used to interpret the correlations among the features and across the time dimension, thereby establishing a human-interpretable way to understand the reconstruction from the model.

Background: Our study hypothesized that an intelligent gradient boosting machine (GBM) model can predict cerebrovascular events and all‐cause mortality in mitral stenosis (MS) with atrial flutter (AFL) by recognizing comorbidities, electrocardiographic and echocardiographic parameters. Methods: The machine learning model was used as a statistical analyzer in recognizing the key risk factors and high‐risk features with either outcome of cerebrovascular events or mortality. Results: A total of 2184 patients with their chart data and imaging studies were included and the GBM analysis demonstrated mitral valve area (MVA), right ventricular systolic pressure, pulmonary artery pressure (PAP), left ventricular ejection fraction (LVEF), New York Heart Association (NYHA) class, and surgery as the most significant predictors of transient ischemic attack (TIA/stroke). MVA, PAP, LVEF, creatinine, hemoglobin, and diastolic blood pressure were predictors for all‐cause mortality. Conclusion: The GBM model assimilates clinical data from all diagnostic modalities and significantly improves risk prediction performance and identification of key variables for the outcome of MS with AFL. [ABSTRACT FROM AUTHOR]

From May 2019, GitHub launched sponsor mechanism indicating that GitHub is moving towards deeper integration of open source development and economic support. It will bring more comprehensive and diversified support to the open source community. However, the number of developers profiting from the sponsor mechanism follows a long tail distribution. Our study found that only 31% of developers who started the sponsor mechanism received rewards, and 39.3% of them only received a reward of one dollar. Our work focuses on identifying what factors affect the availability of sponsorship for developers in open source community. We start by defining 45 features to characterize the developers in four dimensions i.e. Personality, Advertisement, Repository and Behavior. The results of statistical analysis indicate that most of the proposed features differ significantly between the ones who received rewards (short for MTs_Yes) from those that are not. After that, we build machine learning model based on the proposed features to predict MTs_Yes. Compared with the existing work, results show that our method outperforms baselines by 30% for AUC (Area Under the Curve). In addition, we investigated the relative contribution of features in detecting MTs_Yes and analyzed the important features by using an interpretable model SHAP. Finally, based on the experimental results, we put forward corresponding and practical suggestions for developers who want to receive rewards so as to make the community of open source projects develop more harmonious. [ABSTRACT FROM AUTHOR]

Background: Interpretable machine learning (ML) for early detection of cancer has the potential to improve risk assessment and early intervention. Methods: Data from 261 proteins related to inflammation and/or tumor processes in 123 blood samples collected from healthy persons, but of whom a sub‐group later developed squamous cell carcinoma of the oral tongue (SCCOT), were analyzed. Samples from people who developed SCCOT within less than 5 years were classified as tumor‐to‐be and all other samples as tumor‐free. The optimal ML algorithm for feature selection was identified and feature importance computed by the SHapley Additive exPlanations (SHAP) method. Five popular ML algorithms (AdaBoost, Artificial neural networks [ANNs], Decision Tree [DT], eXtreme Gradient Boosting [XGBoost], and Support Vector Machine [SVM]) were applied to establish prediction models, and decisions of the optimal models were interpreted by SHAP. Results: Using the 22 selected features, the SVM prediction model showed the best performance (sensitivity = 0.867, specificity = 0.859, balanced accuracy = 0.863, area under the receiver operating characteristic curve [ROC‐AUC] = 0.924). SHAP analysis revealed that the 22 features rendered varying person‐specific impacts on model decision and the top three contributors to prediction were Interleukin 10 (IL10), TNF Receptor Associated Factor 2 (TRAF2), and Kallikrein Related Peptidase 12 (KLK12). Conclusion: Using multidimensional plasma protein analysis and interpretable ML, we outline a systematic approach for early detection of SCCOT before the appearance of clinical signs. [ABSTRACT FROM AUTHOR]

Traffic incidents pose substantial hazards to public safety and wellbeing, and accurately estimating their duration is pivotal for efficient resource allocation, emergency response, and traffic management. However, existing research often faces limitations in terms of limited datasets, and struggles to achieve satisfactory results in both prediction accuracy and interpretability. This paper established a novel prediction model of traffic incident duration by utilizing a tabular network-TabNet model, while also investigating its interpretability. The study incorporates various novel aspects. It encompasses an extensive temporal and spatial scope by incorporating six years of traffic safety big data from Tianjin, China. The TabNet model aligns well with the tabular incident data, and exhibits a robust predictive performance. The model achieves a mean absolute error (MAE) of 17.04 min and root mean squared error (RMSE) of 22.01 min, which outperforms other alternative models. Furthermore, by leveraging the interpretability of TabNet, the paper ranks the key factors that significantly influence incident duration and conducts further analysis. The findings emphasize that road type, casualties, weather conditions (particularly overcast), and the number of motor and non-motor vehicles are the most influential factors. The result provides valuable insights for traffic authorities, thus improving the efficiency and effectiveness of traffic management strategies. [ABSTRACT FROM AUTHOR]

Introduction: In recent years, the growing interest in building energy consumption and estimation has led to a wealth of energy data and Building Information Modelling (BIM), providing ample opportunities for data-driven algorithms to be widely applied in the building industry. However, despite promising accuracy in data-driven models for building energy estimation, they only consider building elements and their attributes independently and neglect the interconnected relationship of building elements. Also, Current data-driven models lack interpretability and are often treated as black boxes. As a result, the models cannot be fully trusted for engineering without reasoning the underlying mechanisms behind the estimation.Method: This paper emphasizes the potential of graph-based learning algorithms, specifically GraphSAGE, in utilizing the enriched semantic, geometry, and room topology information derived from BIM data. The aim is to identify critical zones within the building based on their energy consumption characteristics. Besides that, the paper proposed a GraphSAGE explainable model by adopting the SHAP with the proposed NE-GraphSAGE prediction model to make more transparency behind the data-driven models.Results and Discussion: Preliminary results demonstrate the potential to improve pre-construction and post-construction steps by identifying critical zones in buildings and identifying the parameters which affected the efficiency of the zones with low energy consumption.

Abstract Background Our study hypothesized that an intelligent gradient boosting machine (GBM) model can predict cerebrovascular events and all‐cause mortality in mitral stenosis (MS) with atrial flutter (AFL) by recognizing comorbidities, electrocardiographic and echocardiographic parameters. Methods The machine learning model was used as a statistical analyzer in recognizing the key risk factors and high‐risk features with either outcome of cerebrovascular events or mortality. Results A total of 2184 patients with their chart data and imaging studies were included and the GBM analysis demonstrated mitral valve area (MVA), right ventricular systolic pressure, pulmonary artery pressure (PAP), left ventricular ejection fraction (LVEF), New York Heart Association (NYHA) class, and surgery as the most significant predictors of transient ischemic attack (TIA/stroke). MVA, PAP, LVEF, creatinine, hemoglobin, and diastolic blood pressure were predictors for all‐cause mortality. Conclusion The GBM model assimilates clinical data from all diagnostic modalities and significantly improves risk prediction performance and identification of key variables for the outcome of MS with AFL.

Introduction: Lung cancer is a prevalent malignancy globally, with approximately 20% of patients developing cardiopulmonary complications after lobectomy. In order to prevent complications, an accurate and personalized method based on machine learning (ML) is required. Methods: During the period of 2017–2021, a retrospective analysis was conducted on the medical records of patients who had undergone lobectomy for non-small cell lung cancer (NSCLC). We performed logical regression, decision tree (DT), random forest (RF), gradient boost DT, and eXtreme gradient boosting analyses to establish an ML model. The ten-fold cross-validation was used to evaluate the performance of multiple ML models based on various evaluation metrics, including accuracy, precision, recall, F1 score, and area under the receiver operating (AUC). Additionally, we also calculated the Kappa value of these model. Each model used grid search to optimize hyper-parameters and then used the interpretability method to provide explanations for the model's Decisions. Results: The study included 718 eligible patients, among whom the incidence of postoperative cardiopulmonary complications was 20.89%. The RF model showed the best comprehensive performance among all models, and its ten-fold cross-validation accuracy, precision, recall, F1 score, and AUC were (OR and 95% confidence interval [CI]) 0.786 (0.738–0.834), 0.803 (0.735–0.872), 0.738 (0.678–0.797), 0.766 (0.714–0.818), 0.856 (0.815–0.898), respectively. The kappa value of the RF model was 0.696 (0.617–0.768). The SHAP method showed that gender, age, and intraoperative blood loss were closely associated with postoperative cardiopulmonary complications. Conclusion: The application of ML methods for predicting postoperative cardiopulmonary complications based on clinical data of patients with NSCLC showed a good performance. The results indicate that ML combined with the SHAP individualized interpretation method has practical clinical value.

Neuropeptides are a diverse and complex class of signaling molecules that regulate a variety of biological processes. Neuropeptides provide many opportunities for the discovery of new drugs and targets for the treatment of a wide range of diseases, and thus, computational tools for the rapid and accurate large-scale identification of neuropeptides are of great significance for peptide research and drug development. Although several machine learning-based prediction tools have been developed, there is room for improvement in the performance and interpretability of the proposed methods. In this work, we developed an interpretable and robust neuropeptide prediction model, named NeuroPred-PLM. First, we employed a language model (ESM) of proteins to obtain semantic representations of neuropeptides, which could reduce the complexity of feature engineering. Next, we adopted a multi-scale convolutional neural network to enhance the local feature representation of neuropeptide embeddings. To make the model interpretable, we proposed a global multi-head attention network that could be used to capture the position-wise contribution to neuropeptide prediction via the attention scores. In addition, NeuroPred-PLM was developed based on our newly constructed NeuroPep 2.0 database. Benchmarks based on the independent test set show that NeuroPred-PLM achieves superior predictive performance compared with other state-of-the-art predictors. For the convenience of researchers, we provide an easy-to-install PyPi package (https://pypi.org/project/NeuroPredPLM/) and a web server (https://huggingface.co/spaces/isyslab/NeuroPred-PLM). [ABSTRACT FROM AUTHOR]

Reynolds-averaged Navier-Stokes (RANS) simulation is currently the primary approach for engineering predictions of turbulent mixing induced by Rayleigh-Taylor (RT), Richtmyer-Meshkov, and Kelvin-Helmholtz instabilities. However, traditional Boussinesq-type RANS mixing models are inadequate for resolving turbulence anisotropy that plays an important role in most of engineering flows. In this study, a data-driven nonlinear K-L mixing model is developed via the gene expression programming (GEP) method to provide an explicit and interpretable model, which can be easily ported to different RANS solvers. Specifically, the Reynolds stress is closed with a second-order truncated generalized Cayley-Hamilton constitutive relation, where the undetermined coefficients are expressed as functions of the Galilean invariants and are trained by the GEP method. Additionally, the realizability principle is considered in the cost function to ensure physics of the flow field. The results of a series of tests confirm that the new model is robust with different mixing problems, though the training is conducted only with the tilted RT mixing problem. When compared with the baseline K-L model, the new model not only significantly improves the predictive accuracy, but also captures physics of turbulence at a higher level. As the model has been explicitly expressed, the improvements are further interpreted by analyzing the equations. To the best of our knowledge, this is the first study to investigate turbulent mixing problems using machine learning methods. [ABSTRACT FROM AUTHOR]