Search

Deep learning is the extension of machine learning technique which attracted the researchers in the recent past and hence the literature is also limited. This research focused on the need for deep learning for effective prediction process, when compared with few other machine learning models. As a proof of concept, a dataset of students’ performance of a school is analysed by various machine learning models. The performance measurements by all such models are compared with deep learning. These are assessed by two familiar statistical measures, Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE). The error rates are low for deep learning when compared with other models. It is been evident that the result obtained by the deep learning model is more effective than the other machine learning models. The results are improved by reducing the error rates at the rates of 35% by mean absolute error and 48% by root mean squared error.

Smart cities are part of the ongoing advances in technology to provide a better life quality to its inhabitants. Urban mobility is one of the most important components of smart cities. Due to the growing number of vehicles in these cities, urban traffic congestion is becoming more common. In addition, finding places to park even in car parks is not easy for drivers who run in circles. Studies have shown that drivers looking for parking spaces contribute up to 30% to traffic congestion. In this context, it is necessary to predict the spaces available to drivers in parking lots where they want to park. We propose in this paper a new system that integrates the IoT and a predictive model based on ensemble methods to optimize the prediction of the availability of parking spaces in smart parking. The tests that we carried out on the Birmingham parking data set allowed to reach a Mean Absolute Error (MAE) of 0.06% on average with the algorithm of Bagging Regression (BR). This results have thus improved the best existing performance by over 6.6% while dramatically reducing system complexity.

High Blood Pressure (BP) is one of the most dangerous and widespread diseases due to which uncontrolled BP increases the risk of health problems that affects many body organs. Unfortunately, accurate BP measurements require several medical devices and a specialist person who has experience in BP measurement and measured in separate times. Thus, this paper proposes a simple calibration-free method to estimate BP by training BP and Photoplethysmography (PPG) data signals on a machine learning (ML) regression model. The proposed method overcomes drawbacks of BP measurement accuracy and provides enough capability for reliable and calibration-free BP estimation. The obtained results clarify that the error standard deviation (STD) is about 5.3 and 6.4 mmHg of systolic pressure (SP) and diastolic pressure (DP), respectively. In addition, the mean absolute error (MAE) is about 4.2 and 4.5 mmHg of SP and DP, respectively. These results achieve grade “A” for both SP and DP based on the Britain Hypertension Society (BHS) standard. Finally, the results of BP estimation regression models meet the International Organization for Standardization (ISO) requirements for non-invasive BP devices and consequently they can be utilized later in life experiments.

Developing an accurate model is extremely important to design efficient proton exchange membrane fuel cells (PEMFCs) systems. The current work proposes the Whale Optimization Algorithm\ud (WOA) for establishing an accurate and reliable PEMFC models. The idea is to increase accuracy\ud of the extracted model parameters by minimizing error between the experimental and estimated polarization curves. WOA is utilized to mainly mitigate the effect of the local optimum stagnation and\ud the premature convergence that appear with most of literature methods applied in this regard. The\ud effectiveness of the WOA in modeling the PEMFC generators is demonstrated by the experiments\ud using Heliocentris FC50 PEMFC test bench. In contrast to the existing works that characterized the\ud behavior of the PEMFCs under fixed temperature values, the performance of developed model in\ud providing accurate results is investigated under different operating conditions. The efficacy of the\ud WOA is further checked using the data of two PEMFCs available in the literature, namely BCS500W, and Ballard V. Moreover, a comparison is done with some challenging literature techniques\ud along with the necessary statistical analysis. The final results prove that the WOA has very competitive performance. The proposed WOA has produced the lowest Mean Absolute Error among all\ud tested methods, with values of 0.0589V, 0.2323V, and 0.2867V for Heliocentris FC 50, BCS, and\ud Ballard fuel cells, respectively. Besides, the constructed Heliocentris FC 50 model yields highly\ud accurate results, especially under varying temperature values. Owing to this, it can be stated that the\ud WOA a powerful modeling tool. Therefore, it is highly recommended to be employed for creating\ud high-quality PEMFC models.

The impact of reinforced concrete beam-column joints on the shear strength of a building under cyclic loading depends on the types of joints applied. This study considers models of the uniaxial and biaxial joint shear strength of exterior beam-column joints. Prediction models of the uniaxial shear strength under uniaxial cyclic loading based on ACI 352, ASCE 41, and gene expression programming (GEP) have been developed. The ACI 352, ASCE 41, and GEP formulas have the potential to achieve improved results. This study considers a means by which to improve the results of previous models through a proposed deep neural network (DNN) model with three hidden layers among the artificial neural network structures. The R-squared value and mean absolute error determined through this DNN model are 97.94% and 34.13% for the uniaxial model and 98.28% and 2.70% for the biaxial model, respectively. These results indicate that the DNN model is more suitable than the ACI 352, ASCE 41, and GEP models for joint shear strength predictions.

Mobility is considered one of the main reasons for the COVID-19 spread. Predicting the effect of control measures on mobility is essential to apply effective decisions. This work proposes an AI-based model for mobility changes prediction. The proposed CNN-LSTM with Autoregression has achieved the best results compared to other investigated models. Results show that the proposed model can predict the effect of precaution control measures on future community mobility with minimum loss. The mean absolute error over all countries in the study is 5.3 . For Egypt and Saudi Arabia, the model achieved an MAE loss of 4.6 and 3.7 consecutively.

Data-driven techniques improve the quality of talent training comprehensively for university by discovering potential academic problems and proposing solutions. We propose an interpretable prediction method for university student academic crisis warning, which consists of K-prototype-based student portrait construction and Catboost–SHAP-based academic achievement prediction. The academic crisis warning experiment is carried out on desensitization multi-source student data of a university. The experimental results show that the proposed method has significant advantages over common machine learning algorithms. In terms of achievement prediction, mean square error (MSE) reaches 24.976, mean absolute error (MAE) reaches 3.551, coefficient of determination ($$R^{2}$$ R 2 ) reaches 80.3%. The student portrait and Catboost–SHAP method are used for visual analysis of the academic achievement factors, which provide intuitive decision support and guidance assistance for education administrators.

In this study, the soft computing technique of Gene expression programming (GEP) has been employed to generate a predictive equation of infiltration rate (fp). Infiltration experiments were conducted at 124 different sites and soil samples were collected to assess various soil properties throughout the Himalayan lake catchment. Parameters determined from observed data using nonlinear-Levenberg Marquardt algorithm were substituted in Horton, Kostiakov and Philip infiltration models and fp were predicted. Using soil data generated by laboratory investigation of soil samples, the GEP model was developed. Training and testing of the GEP model was performed using 70% and 30% of data respectively. Performance of GEP developed functional relationship was evaluated by comparing predictions from it and aforementioned infiltration models with field observed fp, and by applying overall performance index (OPI) computed using Coefficient of Determination (R2), Nash–Sutcliffe Efficiency (ENS), Willmott’s Index of Agreement (W), Mean Absolute Error (MAE) and Root Mean Square Error (RMSE). Expression developed using GEP indicated feasibility of developed equation with ENS, R2, W, RMSE and MAE of 0.84, 0.84, 0.96, 1.9, and 0.8, respectively for training data-set and 0.84, 0.85, 0.95, 1.2, and 0.95, respectively for testing data-set. Comparative analysis revealed that though with a slightly higher OPI value (0.7–0.8), the performance of conventional models is better compared to the GEP model (0.66) but the GEP model having satisfactory performance may be used for fp prediction particularly in absence of observed data.

Dissolved oxygen (DO), an important water quality indicator in aquaculture, affects the survival rate of aquatic creatures and the yield of aquatic production. Therefore, it is important to predict DO in fishery ponds for applying artificial aeration with low energy and cost. Recently, deep learning models, such as recurrent neural network (RNN), long short-term memory (LSTM), and gated recurrent unit (GRU), are often used to predict the trend of time series, but it is unclear which one of them is more suitable for prediction of DO in fishery ponds. In this work, the RNN model, LSTM model, and GRU model were used to build three DO predicting models. The performance of the three models were compared by mean absolute error (MAE), mean square error (MSE), mean absolute percentage error (MAPE), and the coefficient of determination (R2). The performance of RNN is worse result than LSTM and GRU. The four evaluation indicators of GRU are 0.450 mg/L, 0.411, 0.054, and 0.994, and the four indicators of LSTM are 0.407 mg/L, 0.294, 0.059, and 0.970, which shows that the performance of GRU is similar to LSTM, but the time cost and number of parameters used for GRU is much lower than LSTM. It is concluded that the GRU has overall better performance and can be applied to practical applications.

In this study, new hybrid artificial neural network (ANN) models were used for predicting the groundwater resource index. The salp swarm algorithm (SSA), particle swarm optimization (PSO), and genetic algorithm (GA) were used to find the weight and bias values of the ANN models. The ANN-PSO, ANN-SSA and ANN-GA models were used to predict the groundwater resource index (GRI)-based drought at different timescales (6, 12, and 24 months) in Yazd plain, Iran. Five input scenarios were used for modeling GRI. The best input scenario was a combination of one-month-lagged GRI, two-month-lagged GRI, three-month-lagged GRI, four-month-lagged GRI, and five-month-lagged GRI, which is known as the fifth input scenario. The outputs of models indicated that the ANN-SSA model with input scenario (5) decreased the mean absolute error (MAE) of ANN-PSO (5) and ANN-GA (5) by 43% and 51%, respectively. Among the hybrid ANN models, ANN-SSA (5), ANN-PSO (5) and ANN-GA (5) outperformed the other hybrid ANN models.

MODIS land surface temperature (LST) product (MOD11A1) has been widely used in analysing spatiotemporal trends of LST. However, its applicability is limited, partially due to its coarse spatial resolution (i.e., 1 km). In this study, an Adaptive random forest regression (ARFR) method was developed for LST downscaling at national scale. This study also provided a framework to shift from downscaling single-time image sets to extensive time-series of MOD11A1 LST images in an operational approach (i.e., a 19-years spatiotemporal LST trend analysis over Iran) using the Google Earth Engine (GEE) cloud computing platform. The performance of ARFR was assessed by comparing the results of the downscaled LSTs with the Landsat-8 LST data on different dates of six consecutive years (2014-2019) over ten different sub-areas in Iran. The results demonstrated the effectiveness of the proposed method with an average root mean square error and mean absolute error of 2.22 °C and 1.59 °C, respectively. The results of spatiotemporal LST trend analysis showed that 25.08%, 10.05%, 56.68%, 1.04%, and 32.84% of Iran experienced significant positive trends during a full year, spring, summer, fall, and winter, respectively. Significant negative trends were also observed over the 3.09%, 23.84%, 7.54%, 17.38%, and 18.77% of Iran during a full year, spring, summer, fall, and winter, respectively. In summary, the outcomes of this study not only exhibit the spatiotemporal trends of LST across Iran, but also reveal the substantial benefits of the ARFR method in downscaling LST using GEE.

Ingredient amounts are crucial for food-oriented health systems, but this information is seldom used in food-oriented health systems due to the difficulty of fetching it from online recipes. This study proposes a predictive model named DeepRecipes to extract ingredient amounts from online textual recipes. The model predicts ingredient amounts according to a given recipe’s name and listed ingredients. We train the model on a small set of recipes containing all ingredients and their corresponding amounts. As we can extract the recipe names and ingredients from almost all online recipes, the proposed model can potentially recover ingredient amounts for massive online recipes. We first trained the model on a small set of recipes containing all ingredients and their corresponding amounts. Then, we compared ten models as references for their performances. The performance of DeepRecipes exceeds those of all the comparison models. The model’s mean absolute error (MAE) and mean absolute percentage error (MAPE) are $3.96\times {10}^{-1}$ and 18.57%, respectively, and its APEs are lower than 50% in more than 95% of the total predictions. This accuracy is sufficient for providing rough ingredient amount estimations for food-oriented health systems.

The aim of the study was to evaluate evapotranspiration (ET) estimation models for the municipality of Macapá, AP. The models studied were Hargreaves-Samani (HS), Turc, Camargo (CM), Jensen-Haise (JH) and Solar Radiation (RS). The ET estimated by the models were compared with those obtained by the standard Penman-Monteith method. Statistical tests of Willmot's Concordance index (d), Correlation Coefficient (r), Performance coefficient (c), Mean Absolute Error (EAM), Mean Square Error and Residual Mass Coefficient (CMR) were performed. With the results of "c" only HS, in August (c = 0.71; "good") and December (c = 0.73; "good") presented a performance considered good. JH had the worst result, between 0.05 and 0.22, considered terrible. In relation to “d”, HS also had a better result, with 0.79 and 0.83 in August and December. The worst was JH, with a minimum value of 0.09 for May. For CMR, CM, JH and RS overestimated ET. TC, on the other hand, presented both underestimation and overestimation. HS overestimated from January to July and underestimated in the other months. For EAM, the models tended to overestimate ET. For MSE, HS presented the lowest values. JH was the one with the worst results, above 4. And the coefficient of determination showed a small dispersion of data, with R2 above 0.90 for HS, TC, JH and RS. Based on what was exposed, HS was the method that presented the best responses, followed by CT and RS. And JH was the one with the highest overestimation rate.

This paper explores two new weightless neural network models, Regression WiSARD and ClusRegression WiSARD, in the challenging task of predicting the total palm oil production of a set of 28 (twenty eight) differently located sites under different climate and soil profiles. Both models were derived from Kolcz and Allinson’s n-Tuple Regression weightless neural model and obtained mean absolute error (MAE) rates of 0.09097 and 0.09173, respectively. Such results are very competitive with the state-of-the-art (0.07983), whilst being four orders of magnitude faster during the training phase. Additionally the models have been tested on three classic regression datasets, also presenting competitive performance with respect to other models often used in this type of task.

Despite the increased share of photovoltaic (PV) systems, several integration challenges are still present. For small-scale PV, the incentives for monitoring and forecasting the output are generated from the gains of optimizing of local electricity usage. PV output can often be the only measured parameter available from a PV site. Thereby, its efficient utilization in adjusting the PV conversion model is essential. The aim of this study is to develop a straightforward and explicit approach for utilizing PV output data in adjusting a site-specific PV output forecast. A parametric PV output conversion model is utilized as the baseline model for 23 PV sites in Finland, implemented on top of five summer months of day-ahead Numerical Weather Prediction (NWP) forecasts, provided by the MetCoOp system. The proposed approach is validated by comparing the performance of the baseline and adjusted forecasts against actual PV output. The method is shown to provide a solution for adjusting a generic PV output forecast to any specific site, reducing the hourly Mean Absolute Error (MAE) of the aggregated forecast from 0.08 to 0.06 W/Wp, while reducing bias from 0.05 to 0.01 W/Wp, respectively. The adjustment increases modeling accuracy by capturing potential systematic biases of the NWP forecast and site-specific conditions, e.g., variable shadowing and conversion losses, affecting the output of the PV system. Due to the fact that the approach can be implemented both for observations and forecast input data, the method is considered to be a helpful approach for several different PV modeling applications.

$\text{HbA}_{\text{1c}}$ is a primary marker of long-term average blood glucose, which is an essential measure of successful control in type 1 diabetes. Previous studies have shown that $\text{HbA}_{\text{1c}}$ estimates can be obtained from 5-12 weeks of daily blood glucose measurements. However, these methods suffer from accuracy limitations when applied to incomplete data with missing periods of measurements. The aim of this article is to overcome these limitations improving the accuracy and robustness of $\text{HbA}_{\text{1c}}$ prediction from time series of blood glucose. A novel data-driven $\text{HbA}_{\text{1c}}$ prediction model based on deep learning and convolutional neural networks is presented. The model focuses on the extraction of behavioral patterns from sequences of self-monitored blood glucose readings on various temporal scales. Assuming that subjects who share behavioral patterns have also similar capabilities for diabetes control and resulting $\text{HbA}_{\text{1c}}$ , it becomes possible to infer the $\text{HbA}_{\text{1c}}$ of subjects with incomplete data from multiple observations of similar behaviors. Trained and validated on a dataset, containing 1543 real world observation epochs from 759 subjects, the model has achieved the mean absolute error of 4.80 $\pm\; \text{0.62}$ mmol/mol, median absolute error of 3.81 $\pm\; \text{0.58}$ mmol/mol and $\text{R}^2$ of 0.71 $\pm$ 0.09 on average during the 10 fold cross validation. Automatic behavioral characterization via extraction of sequential features by the proposed convolutional neural network structure has significantly improved the accuracy of $\text{HbA}_{\text{1c}}$ prediction compared to the existing methods.

Effective life-cycle management of deteriorating structures should be based on an accurate and reliable damage propagation prediction model under uncertainty. Appropriate updating with inspected in...

In e-commerce, the opinion of users about products and the reviews are identified using recommender systems. Collaborative filtering techniques are popularly used techniques for giving recommendations to the users. One of the common challenges in the collaborative filtering technique for giving recommendations is cold start problem, which occurs due to insufficient information about new items and new users. This paper proposes a hybrid approach entitled LA-ALS to address the cold start problem to provide effective recommendations. The LA-ALS approach makes use of the benefits of both Louvain’s algorithm and alternating least square algorithm. The Louvain’s algorithm is used to analyze the relationship between users and alternating least square algorithm is used to predict recommendations. Experiments are carried out by using real-world datasets such as Movielens and Facebook databases. The effectiveness of the LA-ALS approach is shown with two parameters namely mean absolute error and root mean square error. The results showed that LA-ALS approach generated better recommendations when compared with the existing techniques such as k-nearest neighbors and singular value decomposition.

Accurate estimation of streamflow has a vital importance in water resources engineering, management and planning. In the present study, the abilities of group method of data handling-neural networks (GMDH-NN), dynamic evolving neural-fuzzy inference system (DENFIS) and multivariate adaptive regression spline (MARS) methods are investigated for monthly streamflow prediction. Precipitation, temperature and streamflows from Kalam and Chakdara stations at Swat River basin (mountainous basin), Pakistan, are used as inputs to the applied models in the form of different input scenarios, and models’ performances are evaluated on the basis of root mean square error (RMSE), mean absolute error (MAE), Nash–Sutcliffe efficiency (NSE) and combined accuracy (CA) indexes. Test results of the Kalam Station show that the DENFIS model provides more accurate prediction results in comparison of GMDH-NN and MARS models with the lowest RMSE (18.9 m3/s), MAE (13.1 m3/s), CA (10.6 m3/s) and the highest NSE (0.941). For the Chakdara Station, the MARS outperforms the GMDH-NN and DENFIS models with the lowest RMSE (47.5 m3/s), MAE (31.6 m3/s), CA (26.1 m3/s) and the highest NSE (0.905). Periodicity (month number of the year) effect on models’ accuracies in predicting monthly streamflow is also examined. Obtained results demonstrate that the periodicity improves the models’ accuracies in general but not necessarily in every case. In addition, the results also show that the monthly streamflow could be successfully predicted using only precipitation and temperature variables as inputs.

In this research, a new framework has been introduced for rainfall temporal variability evaluation by using combination of monthly rainfall data sets in three synoptic stations, Principal Component Analysis (PCA), Adaptive Neuro Fuzzy Inference System (ANFIS), Grasshopper Optimization Algorithm (GOA), and Innovative Trend Analysis (ITA) methodology. The first five components were chosen as inputs of the soft-computing models, based on PCA. The GOA was used for training the ANFIS model, in order to estimate the monthly rainfall. The outputs of the ANFIS-GOA were compared to the rainfall estimates by ANFIS-Particle Swarm Optimization (ANFIS-PSO) and ANFIS-Genetic Algorithm (ANFIS-GA). Moreover, various statistical indices, such as mean absolute error (MAE), percent bias (PBIAS) and Nash-Sutcliffe Efficiency (NSE) were used to evaluate the soft-computing models’ performance. Results indicated that ANFIS-GOA had higher accuracy in estimating the rainfall (values of MAE, NSE and PBIAS were 0.21, 0.92 and 0.16 for Mehrabad station, 0.16, 0.94 and 0.14 for Semnan station and 0.24, 0.91 and 0.17 for Noshahr station, respectively) in the testing phase. These values showed significant improvements (67.8%, 21% and 40% for Mehrabad station, 69.2%, 17.5% and 33.3% for Semnan station and 57.1%, 21.3% and 37% for Noshahr station) versus indices related to standalone ANFIS model, which reflected the supremacy and higher accuracy of ANFIS-GOA model in rainfall prediction for different climatic conditions. It was also concluded that the ANFIS-GOA, ANFIS-PSO, and ANFIS-GA models performed superior to the standalone ANFIS-based model, respectively. Furthermore, possible trends in monthly rainfall have been detected by ITA, which is a new graphical model. Results showed significant decreasing trends in January and July for all the rainfall values in Mehrabad station. By comparison of the results obtained from ANFIS and the hybrid models with observed data, it was also concluded that the trends of observed data were close to the ANFIS-GOA predictions.

This research aims to evaluate two econometric models to forecast imports and exports for the financial year (FY) 2020. For this purpose, we used the annual exports and imports data of Pakistan from FY2002 to FY2019. Thus, in this regard, we employed, and compared the results of two econometrics models such as Box Jenkins or Autoregressive Integrated Moving Average (ARIMA), and Auto-Regressive (AR) with seasonal dummies. For examining the precision of forecasting, we employed mean absolute error and root mean square error approaches. The findings of Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) reveal that the ARIMA or Box Jenkins approach provides better accuracy of the forecast for the exports as compared to the AR model with dummies. However, Auto-Regressive (AR) model has demonstrated more precision for the imports as compared to the Box Jenkins model. Hence, the projected forecasting for the growth of export is 1.87% for the FY2020 and projected forecasting for the import demonstrates a negative variation of -1.61% for the FY2020. The findings of the undertaken study recommend the policymakers of Pakistan to take corrective measures to increase exports and to prevent the country from the trade deficit. The policymakers of Pakistan should give more incentives to the exporters and decrease the cost of doing business to be more competitive than the regional economies such as India, Bangladesh, and China.

Disparity map quality assessment is crucial to evaluate the accuracies of stereo matching algorithms. Several widely used measures such as root mean square (RMS), mean absolute error (MAE) and bad ...

The Cheliff watershed has one of the most spatially diverse pluviometric regimes in northwestern Algeria. Understanding these regimes is essential for managing water resources and identifying the most vulnerable regions to climate change. Mean annual rainfall data (1972–2012) for 58 meteorological stations and their corresponding elevation were used. Maps were produced using three geostatistical interpolation algorithms: ordinary kriging (OK), regression-kriging (RK), and kriging with external drift (KED); the first algorithm uses only rainfall while the other two use also elevation. Interpolation methods were compared using statistical indicators of cross-validation. Results indicate that KED is the least biased interpolator with limited number of strong underestimates or overestimates and limited relative importance of this strong underestimation or overestimation, followed by RK and finally OK. The best match between measured and predicted values was for KED (correlation coefficient of 0.82), followed by RK (0.79), while OK is far from them (0.70). KED can be considered the best model because it gives the lowest values of mean error, mean absolute error, and root mean square error (− 1.9, 35.4, and 49.5 mm, respectively) and the highest values of Willmott agreement index, Lin concordance coefficient, and Nash–Sutcliffe efficiency coefficient (0.89, 0.80, and 0.67, respectively), results of RK are intermediate, while those of OK are the worst. There is clearly significant improvement in the prediction performance taking into account the elevation, in particular by KED. Results show that KED is the most appropriate to produce map of mean annual rainfall in the Cheliff watershed, Algeria.

Continuous blood pressure (BP) monitoring is important for patients with hypertension. However, BP measurement with a cuff may be cumbersome for the patient. To overcome this limitation, various studies have suggested cuffless BP estimation models using deep learning algorithms. A generalized model should be considered to decrease the training time, and the model reproducibility should be taken into account in multi-day scenarios. In this study, a BP estimation model with a bidirectional long short-term memory network is proposed. The features are extracted from the electrocardiogram, photoplethysmogram, and ballistocardiogram. The leave-one-subject-out (LOSO) method is incorporated to generalize the model and fine-tuning is applied. The model was evaluated using one-day and multi-day tests. The proposed model achieved a mean absolute error (MAE) of 2.56 and 2.05 mmHg for the systolic and diastolic BP (SBP and DBP), respectively, in the one-day test. Moreover, the results demonstrated that the LOSO method with fine-tuning was more compatible in the multi-day test. The MAE values of the model were 5.82 and 5.24 mmHg for the SBP and DBP, respectively.

We propose a novel deep learning method for shadow removal. Inspired by physical models of shadow formation, we use a linear illumination transformation to model the shadow effects in the image that allows the shadow image to be expressed as a combination of the shadow-free image, the shadow parameters, and a matte layer. We use two deep networks, namely SP-Net and M-Net, to predict the shadow parameters and the shadow matte respectively. This system allows us to remove the shadow effects from images. We then employ an inpainting network, I-Net, to further refine the results. We train and test our framework on the most challenging shadow removal dataset (ISTD). Our method improves the state-of-the-art in terms of root mean square error (RMSE) for the shadow area by 20\%. Furthermore, this decomposition allows us to formulate a patch-based weakly-supervised shadow removal method. This model can be trained without any shadow-free images (that are cumbersome to acquire) and achieves competitive shadow removal results compared to state-of-the-art methods that are trained with fully paired shadow and shadow-free images. Last, we introduce SBU-Timelapse, a video shadow removal dataset for evaluating shadow removal methods., Comment: PAMI21 - Camera Ready Version. arXiv admin note: substantial text overlap with arXiv:1908.08628

Accurate corrosion predictions are vital to safe and optimised designs of marine assets. Traditional approaches, including those used to develop rule requirements, seek to use empirical regressions to model corrosion, but most are solely time-dependent. This may lead to conservative damage estimates and hence heavy and inefficient ships. To provide more accurate predictions, this paper presents an interpretable machine learning algorithm based on data fusion of ship survey and experimental measurements. The corrosion behaviour in bulk carrier ballast tanks is interpreted through a sensitivity analysis which quantifies the relationships between operational/environmental factors and the corrosion rate. The prediction accuracy is improved by a minimum of 82% when compared to the two representative empirical models, with a mean absolute error down to 0.10 mm.

Currently, renewable-energy-based power generation is rapidly developing to tackle climate change; however, the use of renewable energy is limited owing to the uncertainty related to renewable energy sources. In particular, energy storage systems (ESSs), which are critical for implementing wind power generation (WPG), entail a wide uncertainty range. Herein, a reinforcement leaning (RL)-based ESS operation strategy is investigated for managing the WPG forecast uncertainty. First, a WPG forecast uncertainty minimization problem is formulated with respect to the ESS operation, subject to ESS constraints, and then, the problem is presented as a Markov decision process (MDP) model, with the state-action space limited by the ESS characteristics. To achieve the optimal solution of the MDP model, an expected state-action-reward-state-action (SARSA) method, which is robust toward the dispersion of the system environment, is employed. Further, frequency-domain data screening based on the k-mean clustering method is implemented to improve learning performance by reducing the variance of the WPG forecast uncertainty. Extensive simulations are conducted based on practical WPG generation data and forecasting. Results indicate that the proposed clustered RL-based ESS operation strategy can manage the WPG forecast uncertainty more effectively than conventional Q-learning-based methods; additionally, the proposed method demonstrates a near-optimal performance within a 1%-point analysis gap to the optimal solution, which requires complete information, including future values.

This paper presents a novel approach for continuous gait phase estimation for human level walking, stair ascent and stair descent relying only on the kinematic variables of the shank, which are measurable by a single Inertial Measurement Unit (IMU) placed at the shank. We use data from an experiment with an instrumented stair to train Artificial Neural Networks (ANNs) and to obtain the data necessary for a k-Nearest-Neighbour (kNN) method. Both methods are used for a continuous gait phase estimation separately for each of the three locomotion modes level walking, stair ascent and stair descent. The so called pseudo- velocities are introduced, a substitution for anslational velocities as input values. The presented gait phase estimation with ANNs achieves a good performance (mean absolute error > 6%) for all three locomotion modes for one test subject and is much faster in comparison to a kNN approach. The use of ANNs seams promising regarding performance and speed for a future implementation on an active prosthesis.

This article introduces fiber Bragg grating (FBG) system for monitoring selected traffic flow parameters in urban areas. The system is able to monitor traffic density or speed of cars driving in single lane. The proposed system consists of five Bragg gratings encapsulated in Polydimethylsiloxane polymer and is characterized by easy and inexpensive installation in the milled upper layer of the road covered with an asphalt mixture. The results of first operational tests presented in this study include 3978 passing vehicles of various specifications. The system's vehicle detection success rate is 99.62%. And further, a mean absolute error 1.35 kph and a relative error 2.62% when measuring vehicle speed. Web of Science 8 127658 127648

Acquired Immune Deficiency Syndrome (AIDS) is still one of the most life-threatening diseases in the world. Moreover, new infections are still potentially increasing. This difficult problem must be solved. Early warning is the most effective way to solve this problem. Here, we aim to determine the best performing model to track the epidemic of AIDS, which will provide a methodological basis for testing the time characteristics of the disease. From January 2004 to January 2018, we built four computing methods based on AIDS dataset: BPNN model, RNN model, LSTM model and MHPSO-GRU model. Compare the final estimated performance to determine the preferred method. Result. Considering the root mean square error (RMSE), mean absolute error (MAE), mean error rate (MER) and mean absolute percentage error (MAPE) in the simulation and prediction subsets, the MHPSO-GRU model is determined as the best performance technology. Estimates for the period from May 2018 to December 2020 suggest that the event appears to continue to increase and remain high., This paper uses RNN derived network GRU to predict the incidence rate of AIDS. The mhpso-gru network prediction model of HIV incidence sequence was established. Mhpso Gru network optimizes the initial weight of deep Gru network through mhpso, and then restores the optimal solution to the initial weight of deep Gru network.

Bayesian and discriminant function analysis (DFA) models have recently been used as tools to estimate sediment source contributions. Unlike existing multivariate mixing models, the accuracy of these two models remains unclear. In the current study, four well-distinguished source samples were used to create artificial mixtures to test the performance of Bayesian and DFA models. These models were tested against the Walling-Collins model, a credible model used in estimation of sediment source contributions estimation, as a reference. The artificial mixtures were divided into five groups, with each group consisting of five samples with known source percentages. The relative contributions of the sediment sources to the individual and grouped samples were calculated using each of the models. The mean absolute error (MAE) and standard error of (SE) MAE were used to test the accuracy of each model and the robustness of the optimized solutions. For the individual sediment samples, the calculated source contributions obtained with the Bayesian (MAE = 7.4%, SE = 0.6%) and Walling-Collins (MAE = 7.5%, SE = 0.7%) models produced results which were closest to the actual percentages of the source contributions to the sediment mixtures. The DFA model produced the worst estimates (MAE = 18.4%, SE = 1.4%). For the grouped sediment samples, the Walling-Collins model (MAE = 5.4%) was the best predictor, closely followed by the Bayesian model (MAE = 5.9%). The results obtained with the DFA model were similar to the values for the individual sediment samples, with the accuracy of the source contribution value being the poorest obtained with any of the models (MAE = 18.5%). An increase in sample size improved the accuracies of the Walling-Collins and Bayesian models, but the DFA model produced similarly inaccurate results for both the individual and grouped sediment samples. Generally, the accuracy of the Walling-Collins and Bayesian models was similar (p > 0.01), while there were significant differences (p

In this paper, a new approximation for the standard normal cumulative distribution function, Φ(z) is proposed. The maximum absolute error (MAE) of the proposed approximation is 9.37935 × 10−6. Comp...

Ride-hailing applications have been offering convenient ride services for people in need. However, such applications still suffer from the issue of supply-demand disequilibrium, which is a typical problem for traditional taxi services. With effective predictions on passenger demands, we can alleviate the disequilibrium by pre-dispatching, dynamic pricing or avoiding dispatching cars to zero-demand areas. Existing studies of demand predictions mainly utilize limited data sources, trajectory data, or orders of ride services or both of them, which also lacks a multi-perspective consideration. In this article, we present a unified framework with a new combined model and a road-network-based spatial partition to leverage multi-source data and model the passenger demands from temporal, spatial, and zero-demand-area perspectives. In addition, our framework realizes offline training and online predicting, which can satisfy the real-time requirement more easily. We analyze and evaluate the performance of our combined model using the actual operational data from UCAR. The experimental results indicate that our model outperforms baselines on both Mean Absolute Error and Root Mean Square Error on average.

Laboratory experiments were used to estimate the hydraulic performance of emitters, i.e., the emitter flow variation (qvar) and manufacturer’s coefficient of variation (CVm), by measuring the discharge of different labyrinth-channel emitters at different operating pressures (P) and water temperatures (T). Considering the importance of the structural parameters of the labyrinth-channel emitters in drip irrigation design, which has been experimentally confirmed, artificial neural network (ANN) and gene expression programming (GEP) models were developed to predict qvar and CVm. The ANN and GEP models were trained and tested using structural parameters (including the number, height (H), and spacing of trapezoidal units and the flow path width and length) of different labyrinth-channel emitters, P and T as the input variables, and qvar and CVm as the outputs. Statistical criteria, including the coefficients of correlation (r), relative root-mean-square error (RRMSE), and mean absolute error (MAE), were used to examine the accuracy of the developed models. The ANN models exhibited good correlation with experimental values, with high r values 0.995 and 0.969 for qvar and 0.997 and 0.947 for CVm in the training and testing processes, respectively. The ANN models had lower RRMSE and MAE values than the GEP models. Furthermore, H was the dominant variable for obtaining the most accurate prediction model. The results confirm that the ANN models are superior to the GEP models for the prediction of the hydraulic performance of emitters.

Forecasting the yield of each product is critical for a semiconductor manufacturer. To enhance the performance of forecasting the yield of a semiconductor product, a hybrid-aggregation and entropy-consensus fuzzy collaborative intelligence (FCI) approach is proposed in this study. The novelty of the proposed approach is in its use of a hybrid aggregation mechanism that first aggregates fuzzy yield forecasts by using a fuzzy weighted average (FWA) and then adjusts the FWA result by using fuzzy intersection (FI). In this way, both subjective and objective viewpoints are considered in forecasting the yield of a semiconductor product. In addition, the consensus among experts is measured with the entropy of the aggregation result. After consensus is reached, the aggregation result is defuzzified using a back propagation network (BPN). The effectiveness of the proposed methodology is validated by analyzing a real case. According to the analysis results, the forecasting accuracy, measured in terms of mean absolute error (MAE) or mean absolute percentage error (MAPE), improved considerably when using the proposed methodology.

Due to rise in solar energy applications, research in Photovoltaic (PV) systems is increasingly taking precedence. Ideally, one requires an accurate model of the PV based cell/module so as to theoretically investigate the performances of various control schemes. In this work, a novel hybridized optimizer, hybridized interior search algorithm (HISA), has been explored for effective parameters estimation of PV cell/module. Single and double diode-based models of PV cell/module have been estimated from their respective single I - V nonlinear characteristic using experimental data from five case studies comprising of two cells and three modules using mono-crystalline, multi-crystalline and thin film-based PV technology. Additionally, based on used model and PV technology specific insights on new bounds for parameter estimation are investigated. Modeling performance of HISA has been assessed using root mean square error (RMSE), weighted RMSE and mean absolute error, between computed and experimental data. A comparative study with the 34 recent published works, in reported five case studies including analytical, deterministic and metaheuristics-based methods, revealed that HISA based parameter estimation is superior. Therefore, based on the presented detailed investigation, it is concluded that the proposed HISA is a promising optimization technique for PV cell/module estimation.

This article introduces some Liu parameters in the linear regression model based on the work of Shukur, Mansson, and Sjolander. These methods of estimating the Liu parameter d increase the efficien...

Aim. In the present investigation artificial neural network (ANN) and ARIMA-model are compared for forecasting of data of colour of water.Methods. Data corresponds to the colour of water of groundwater and drinking water of water intake of south-east region of the Republic of Belarus. The definition of colour was carried out for the period from 2009 to 2017. twice a day, the time series of values included 5215 values. The parameters of the models were estimated by 85% of the time series values, and the remaining 15% of the values (the test period) compared the forecast values with the actual ones. Optimal configurations of ARIMA-models were determined from the results of comparing the averaged values of the root mean squared errors (RMSE); optimal configurations of ANN were determined from the results of comparing the averaged values of RMSE and correlation coefficients (CC) on the test periods.Results. Comparison of forecasting methods was carried out on the basis of the averaged values of mean absolute error and mean relative error on the test periods. It was revealed that ANN allows to obtain the predicted values of colour of water more accurate than ARIMA-model.Main conclusions. Software implementation of ANN in the MATLAB environment empowers with sufficient accuracy get forecast values of groundwater and drinking water for 100 values.

This article proposes nine new approximations for the standard normal cumulative distribution function Φz. In addition, it collects most of the approximations existing in the literature. The accura...

To predict the critical heat flux (CHF) of flow boiling under subcooled conditions, this study comes up with a new model based on the Interfacial Lift-off CHF model. The new model makes up the shortcomings of the previous one used only under saturated conditions, and it also considers the effect of condensation on the interface between liquid and vapor phases. An all-new subcooled correctional coefficient is risen up based on the CHF value for the corresponding flow velocity under near-saturated conditions. By comparison between the predicted results of this model and the experimental ones at 6 subcoolings from 3 to 30 °C, and 18 flow velocities from 0.75 to 8 m/s, with the mean absolute error of 9.4%, the predicted trends are both intuitively reasonable and numerically accurate. It reflects the rule under which CHF changes with the outlet subcooling and flow velocity under both saturated and subcooled conditions, and the vapor amount within a vapor bulk decreases with the increase of subcooling. But as the flow velocity increases, it decreases first, and then tends to be a fixed value gradually.

Extra virgin olive oil (EVOO), the star of the Mediterranean diet, is a valuable product within the food sector. In certain cases, due to specific and special olive varietals and blends, as well as organoleptic properties, EVOOs are even authenticated with protected designation of origin (PDO) labels. Because of this, false labeling and even adulterations have been reported in this sector. During this research, a laser diode has been used to gather 254 fluorescence spectra from that many artificially prepared samples of fresh PDO EVOO mixed with small amounts of old olive oils (adulterants). The database was then treated and employed to train and optimize an intelligent model based on a supervised artificial neural network known as multilayer perceptron (MLP). The goal of this non-linear model was dual: to identify the PDO of every sample and to quantify the amount of fresh EVOO. After meticulous validation using blinded samples to assess the performance of the MLP, the outcome was successful. The algorithm could perfectly distinguish the PDOs and determine the amount of fresh EVOO with a mean absolute error of around 1.5% (w/w). Therefore, the combination of a laser diode and cognitive modeling leads to a fast and cost-effective tool able to authenticate PDO labels of EVOOs and estimate the amount of potential adulterating agents such as olive oils from old harvests.

Rooftop photovoltaic (PV) systems are usually behind the meter and invisible to utilities and retailers and, thus, their power generation is not monitored. If a number of rooftop PV systems are installed, it transforms the net load pattern in power systems. Moreover, not only generation but also PV capacity information is invisible due to unauthorized PV installations, causing inaccuracies in regional PV generation forecasting. This study proposes a regional rooftop PV generation forecasting methodology by adding unauthorized PV capacity estimation. PV capacity estimation consists of two steps: detection of unauthorized PV generation and estimation capacity of detected PV. Finally, regional rooftop PV generation is predicted by considering unauthorized PV capacity through the support vector regression (SVR) and upscaling method. The results from a case study show that compared with estimation without unauthorized PV capacity, the proposed methodology reduces the normalized root mean square error (nRMSE) by 5.41% and the normalized mean absolute error (nMAE) by 2.95%, It can be concluded that regional rooftop PV generation forecasting accuracy is improved.

The early detection of the coronavirus disease 2019 (COVID-19) outbreak is important to save people’s lives and restart the economy quickly and safely. People’s social behavior, reflected in their mobility data, plays a major role in spreading the disease. Therefore, we used the daily mobility data aggregated at the county level beside COVID-19 statistics and demographic information for short-term forecasting of COVID-19 outbreaks in the United States. The daily data are fed to a deep learning model based on Long Short-Term Memory (LSTM) to predict the accumulated number of COVID-19 cases in the next two weeks. A significant average correlation was achieved (r=0.83 (p = 0.005)) between the model predicted and actual accumulated cases in the interval from August 1, 2020 until January 22, 2021. The model predictions hadr> 0.7 for 87% of the counties across the United States. A lower correlation was reported for the counties with total cases of https://github.com/Murtadha44/covid-19-spread-risk.

Regression analysis makes up a large part of supervised machine learning, and consists of the prediction of a continuous independent target from a set of other predictor variables. The difference between binary classification and regression is in the target range: in binary classification, the target can have only two values (usually encoded as 0 and 1), while in regression the target can have multiple values. Even if regression analysis has been employed in a huge number of machine learning studies, no consensus has been reached on a single, unified, standard metric to assess the results of the regression itself. Many studies employ the mean square error (MSE) and its rooted variant (RMSE), or the mean absolute error (MAE) and its percentage variant (MAPE). Although useful, these rates share a common drawback: since their values can range between zero and +infinity, a single value of them does not say much about the performance of the regression with respect to the distribution of the ground truth elements. In this study, we focus on two rates that actually generate a high score only if the majority of the elements of a ground truth group has been correctly predicted: the coefficient of determination (also known as R-squared or R2) and the symmetric mean absolute percentage error (SMAPE). After showing their mathematical properties, we report a comparison between R2 and SMAPE in several use cases and in two real medical scenarios. Our results demonstrate that the coefficient of determination (R-squared) is more informative and truthful than SMAPE, and does not have the interpretability limitations of MSE, RMSE, MAE and MAPE. We therefore suggest the usage of R-squared as standard metric to evaluate regression analyses in any scientific domain.

Objective: Continuous glucose monitoring (CGM) enables prediction of the future glucose concentration (GC) trajectory for making informed diabetes management decisions. The glucose concentration values are affected by various physiological and metabolic variations, such as physical activity (PA) and acute psychological stress (APS), in addition to meals and insulin. In this work, we extend our adaptive glucose modeling framework to incorporate the effects of PA and APS on the GC predictions. Methods: A wristband conducive of use by free-living ambulatory people is used. The measured physiological variables are analyzed to generate new quantifiable input features for PA and APS. Machine learning techniques estimate the type and intensity of the PA and APS when they occur individually and concurrently. Variables quantifying the characteristics of both PA and APS are integrated as exogenous inputs in an adaptive system identification technique for enhancing the accuracy of GC predictions. Data from clinical experiments illustrate the improvement in GC prediction accuracy. Results: The average mean absolute error (MAE) of one-hour-ahead GC predictions with testing data decreases from 35.1 to 31.9 mg/dL (p-value = 0.01) with the inclusion of PA information, and it decreases from 16.9 to 14.2 mg/dL (p-value = 0.006) with the inclusion of PA and APS information. Conclusion: The first-ever glucose prediction model is developed that incorporates measures of physical activity and acute psychological stress to improve GC prediction accuracy. Significance: Modeling the effects of physical activity and acute psychological stress on glucose concentration values will improve diabetes management and enable informed meal, activity and insulin dosing decisions.

Abstract The present study aims to compare SARIMA and Holt–Winters model forecasts of mean monthly flow at the V Aniversario basin, western Cuba. Model selection and model assessment are carried out with a rolling cross-validation scheme using mean monthly flow observations from the period 1971–1990. Model performance is analyzed in one- and two-year forecast lead times, and comparisons are made based on mean squared error, root mean squared error, mean absolute error and the Nash–Sutcliffe efficiency; all these statistics are computed from observed and simulated time series at the outlet of the basin. The major findings show that Holt–Winters models had better performance in reproducing the mean series seasonality when the training observations were insufficient, while for longer training subsets, both models were equally competitive in forecasting one year ahead. SARIMA models were found to be more reliable for longer lead-time forecasts, and their limitations after being trained on short observation periods are due to overfitting problems. Article Highlights Comparison based on rolling cross-validation revealed the models forecasts sensibility to available observations amount. HW and SARIMA models perform better when limited observations or long-view forecasting, respectively, otherwise they do similar. HW models were superior modeling less variable monthly flows while SARIMA models better forecast the highly variable periods.

In recent years, more and more recommendation algorithm considers social information. However, in the existing social recommendation algorithm focused on the rating prediction task, the user's rating bias and the item's rating bias are often not considered. Besides, the existing social recommendation algorithms usually fail to deal with the sparse problem of social information. Concerning the problem of ignores rating bias and social information sparsity in the social recommendation, this paper proposes a novel recommendation model that combines implicit information and rating bias based on the traditional matrix factorization model. The social information used includes explicit social information and implicit social information generated by collaborative user network embedding methods. Experimental results on two real public datasets show that compared with other recommendation models, the proposed model reduces 1.13% to 10.09% and 3.14% to 8.38% respectively in the Root Mean Square Error (RMSE) and the Mean Absolute Error (MAE). It shows that our proposed recommendation algorithm has a better recommendation effect.

The interest in contactless or remote heart rate measurement has been steadily growing in healthcare and sports applications. Contactless methods involve the utilization of a video camera and image processing algorithms. Recently, deep learning methods have been used to improve the performance of conventional contactless methods for heart rate measurement. After providing a review of the related literature, a comparison of the deep learning methods whose codes are publicly available is conducted in this paper. The public domain UBFC dataset is used to compare the performance of these deep learning methods for heart rate measurement. The results obtained show that the deep learning method PhysNet generates the best heart rate measurement outcome among these methods, with a mean absolute error value of 2.57 beats per minute and a mean square error value of 7.56 beats per minute.

Educational facilities hold a higher degree of uncertainty in predicting maintenance and repair costs than other types of facilities. Moreover, achieving accurate and reliable maintenance and repair costs is essential, yet very little is known about a holistic approach to learning them by incorporating multi-contextual factors that affect maintenance and repair costs. This study fills this knowledge gap by modeling and validating deep neural networks to efficiently and accurately learn maintenance and repair costs, drawing on 1213 high-confidence data points. The developed model learns and generalizes claim payout records on the maintenance and repair costs from sets of facility asset information, geographic profiles, natural hazard records, and other causes of financial losses. The robustness of the developed model was tested and validated by measuring the root mean square error and mean absolute error values. This study attempted to propose an analytical modeling framework that can accurately learn various factors, significantly affecting the maintenance and repair costs of educational facilities. The proposed approach can contribute to the existing body of knowledge, serving as a reference for the facilities management of other functional types of facilities.

Despite lakes being a key part of the global water cycle and a crucial water resource, there is limited understanding of whether regional or lake-specific factors control water storage variations in small lakes. Here, we study groups of small, unregulated lakes in North Carolina, Washington, Illinois, and Wisconsin, USA using lake level measurements gathered by citizen scientists and lake surface area measurements from optical satellite imagery. We show the lake level measurements to be highly accurate when compared to automated gauges (mean absolute error = 1.6 cm). We compare variations in lake water storage between pairs of lakes within these four states. On average, water storage variations in lake pairs across all study regions are moderately positively correlated (ρ = 0.49) with substantial spread in the degree of correlation. The distance between lake pairs and the extent to which their changes in volume are correlated show a weak but statistically significant negative relationship. Our results indicate that, on regional scales, distance is not a primary factor governing lake water storage patterns, which suggests that other, perhaps lakes-specific, factors must also play important roles.