Statistical Machine Learning Models Research Articles

Application of a Poisson deep neural network model for the prediction of count data in genome-based prediction.

Genomic selection (GS) is revolutionizing conventional ways of developing new plants and animals. However, because it is a predictive methodology, GS strongly depends on statistical and machine learning to perform these predictions. For continuous outcomes, more models are available for GS. Unfortunately, for count data outcomes, there are few efficient statistical machine learning models for large datasets or for datasets with fewer observations than independent variables. For this reason, in this paper, we applied the univariate version of the Poisson deep neural network (PDNN) proposed earlier for genomic predictions of count data. The model was implemented with (a) the negative log-likelihood of Poisson distribution as the loss function, (b) the rectified linear activation unit as the activation function in hidden layers, and (c) the exponential activation function in the output layer. The advantage of the PDNN model is that it captures complex patterns in the data by implementing many nonlinear transformations in the hidden layers. Moreover, since it was implemented in Tensorflow as the back-end, and in Keras as the front-end, the model can be applied to moderate and large datasets, which is a significant advantage over previous GS models for count data. The PDNN model was compared with deep learning models with continuous outcomes, conventional generalized Poisson regression models, and conventional Bayesian regression methods. We found that the PDNN model outperformed the Bayesian regression and generalized Poisson regression methods in terms of prediction accuracy, although it was not better than the conventional deep neural network with continuous outcomes.

Geographically and temporally weighted neural network for winter wheat yield prediction

Accurate prediction of crop yield is essential for agricultural trading, market risk management and food security. Although various statistical models and machine learning models have been developed to enhance prediction accuracy, spatial and temporal non-stationarity, an intrinsic attribute of many geographical processes, is still rarely considered in crop yield modeling. From a statistical point of view, this study respectively provided evidence for the existence of spatial non-stationarity and temporal non-stationarity in winter wheat yield prediction based on geographically weighted regression (GWR) and temporally weighted regression (TWR). Then, a geographically and temporally weighted neural network (GTWNN) model was proposed by integrating artificial neural network (ANN) into geographically and temporally weighted regression (GTWR) using publicly available data sources, including satellite imagery and climate data. For a more credible evaluation, the leave-one-year-out strategy was adopted to make out-of-sample prediction resulting in a total of 12 test years from 2008 to 2019. The experiment results showed that the proposed GTWNN outperformed ANN, GTWR and support vector regression (SVR) achieving the average coefficient of determination (R2) values of 0.766, 0.759 and 0.720 at the three prediction times of end of July, end of June and end of May. Moreover, an extended Moran's I was adopted to assess the degree of spatiotemporal autocorrelation of the prediction errors. The error aggregation of GTWNN was lower than other models, indicating that GTWNN is applicable to addressing spatial non-stationarity in modeling the relationship between predictors and yield response. The methodology proposed in this paper can be extended to handle spatiotemporal non-stationarity in other crop yield predictions and even other environmental phenomena.

A new dam structural response estimation paradigm powered by deep learning and transfer learning techniques

With the rapid development of information and communication techniques, dam structural health assessment based on data collected from structural health monitoring systems has become a trend. This allows for applying data-driven methods for dam safety analysis. However, data-driven models in most related literature are statistical and shallow machine learning models, which cannot capture the time series patterns or learn from long-term dependencies of dam structural response time series. Furthermore, the effectiveness and applicability of these models are only validated in a small data set and part of monitoring points in a dam structural health monitoring system. To address the problems, this article proposes a new modeling paradigm based on various deep learning and transfer learning techniques. The paradigm utilizes one-dimensional convolutional neural networks to extract the inherent features from dam structural response–related environmental quantity monitoring data. Then bidirectional gated recurrent unit with a self-attention mechanism is used to learn from long-term dependencies, and transfer learning is utilized to transfer knowledge learned from the typical monitoring point to the others. The proposed paradigm integrates the powerful modeling capability of deep learning networks and the flexible transferability of transfer learning. Rather than traditional models that rely on experience for feature selection, the proposed deep learning–based paradigm directly utilizes environmental monitoring time series as inputs to accurately estimate dam structural response changes. A high arch dam in long-term service is selected as the case study, and three monitoring items, including dam displacement, crack opening displacement, and seepage are used as the research objects. The experimental results show that the proposed paradigm outperforms conventional and shallow machine learning–based methods in all 41 tested monitoring points, which indicates that the proposed paradigm is capable of dealing with dam structural response estimation with high accuracy and robustness.

P143 The effect of JAK inhibition on the metabolome of neutrophils from patients with rheumatoid arthritis

Abstract Background/Aims Janus kinase inhibitors (JAKi) including baricitinib and tofacitinib target the JAK/STAT pathway and are clinically effective in treating rheumatoid arthritis (RA). Neutrophils play a key role in the pathophysiology of RA and their function is directly affected by these drugs as shown by clinical trials of JAKi which report increased infection rates and transient neutropenia during therapy. The aim of this work is to determine difference in the metabolome of neutrophils from healthy controls and patients with RA, and the effect of JAKi on the metabolome of healthy and RA neutrophils. Methods Neutrophils were isolated from healthy controls (HC) (n = 10) and patients with RA (n = 10 DMARD-naïve, n = 10 Biologics-naive) and pre-incubated with baricitinib, tofacitinib or a pan-JAK inhibitor (all 200ng/mL) for 2h before snap-freezing. Metabolites were extracted by 50:50 v/v AcN:H2O and analysed by 1H NMR spectroscopy using a 700 MHz Avance IIIHD Bruker NMR spectrometer equipped with a TCI cryoprobe. Chenomx, Bruker TopSpin and tameNMR software were used to identify metabolites and process spectra. Statistical analysis and machine learning models were implemented using R and the Mixomics package. Pathway analysis was carried out using Ingenuity (IPA). ROS production in response to 15 min fMLP stimulation was measured using DHR-123 and flow cytometry. Results At 0h, changes in levels of NADP+, NADPH and hypoxanthine in RA neutrophils indicated increased production of ROS via NADPH oxidase and xanthine oxidase compared to HC. Lower levels of glucose and increased levels of lactic acid also suggested increased glycolysis in RA neutrophils compared to HC. However after 2h incubation with JAKi, a difference in the response to these drugs was observed. 80 metabolites were higher in RA neutrophils compared to HC following 2h JAKi treatment. IPA analysis revealed the metabolites higher in RA neutrophils were involved in NAD biosynthesis, NAD phosphorylation/dephosphorylation, amino acid synthesis and transport, and carbohydrate metabolism. Changes in the levels of NADP+, NADPH and hypoxanthine in RA neutrophils indicate that JAKi inhibit ROS production. We therefore confirmed experimentally that JAKi inhibit ROS production in response to fMLP (p < 0.05). IPA also predicted altered neutrophil cell functions including cell death and cell-to-cell signalling. We previously showed that JAKi did not affect the constitutive rate of neutrophil apoptosis, however JAKi significantly inhibit the delay in apoptosis induced by GM-CSF (p < 0.05). No significant difference in the metabolome was observed when comparing tofacinib-treatment and baricitinib-treatment in either the HC or RA patient group. Conclusion The metabolome of RA and HC neutrophils is different at baseline. JAKi alter the levels of intracellular metabolites involved in neutrophil activation and ROS production in RA. This may explain, in part, the efficacy of JAKi in decreasing disease activity and joint damage in RA. Disclosure M. Fresneda Alarcon: None. R. Grosman: None. S. Chokesuwattanaskul: None. M. Phelan: None. H. Wright: None.

Interpretable machine learning approach for predicting COVID-19 risk status of an individual

This study aimed to ascertained using Statistical feature selection methods and interpretable Machine learning models, the best features that predict risk status (“Low”, “Medium”, “High”) to COVID-19 infection. This study utilizes a publicly available dataset obtained via; online web-based risk assessment calculator to ascertain the risk status of COVID-19 infection. 57 out of 59 features were first filtered for multicollinearity using the Pearson correlation coefficient and further shrunk to 55 features by the LASSO GLM approach. SMOTE resampling technique was used to incur the problem of imbalanced class distribution. The interpretable ML algorithms were implored during the classification phase. The best classifier predictions were saved as a new instance and perturbed using a single Decision tree classifier. To further build trust and explainability of the best model, the XGBoost classifier was used as a global surrogate model to train predictions of the best model. The XGBoost individual’s explanation was done using the SHAP explainable AI-framework. Random Forest classifier with a validation accuracy score of 96.35 % from 55 features reduced by feature selection emerged as the best classifier model. The decision tree classifier approximated the best classifier correctly with a prediction accuracy score of 92.23 % and Matthew’s correlation coefficient of 0.8960. The XGBoost classifier approximated the best classifier model with a prediction score of 99.7 %. This study identified COVID-19 positive, COVID-19 contacts, COVID-19 symptoms, Health workers, and Public transport count as the five most consistent features that predict an individual’s risk exposure to COVID-19.

Impact of environmental factors in predicting daily severity scores of atopic dermatitis.

BackgroundAtopic dermatitis (AD) is a chronic inflammatory skin disease that affects 20% of children worldwide. Environmental factors including weather and air pollutants have been shown to be associated with AD symptoms. However, the time‐dependent nature of such a relationship has not been adequately investigated. This paper aims to assess whether real‐time data on weather and air pollutants can make short‐term prediction of AD severity scores.MethodsUsing longitudinal data from a published panel study of 177 paediatric patients followed up daily for 17 months, we developed a statistical machine learning model to predict daily AD severity scores for individual study participants. Exposures consisted of daily meteorological variables and concentrations of air pollutants, and outcomes were daily recordings of scores for six AD signs. We developed a mixed‐effect autoregressive ordinal logistic regression model, validated it in a forward‐chaining setting and evaluated the effects of the environmental factors on the predictive performance.ResultsOur model successfully made daily prediction of the AD severity scores, and the predictive performance was not improved by the addition of measured environmental factors. Potential short‐term influence of environmental exposures on daily AD severity scores was outweighed by the underlying persistence of preceding scores.ConclusionsOur data does not offer enough evidence to support a claim that weather or air pollutants can make short‐term prediction of AD signs. Inferences about the magnitude of the effect of environmental factors on AD severity scores require consideration of their time‐dependent dynamic nature.

Pattern similarity-based machine learning methods for mid-term load forecasting: A comparative study

Pattern similarity-based frameworks are widely used for classification and regression problems. Repeated, similar-shaped cycles observed in seasonal time series encourage the use of such frameworks for forecasting. In this paper, we use pattern similarity-based models for mid-term load forecasting. An integral part of these models is the use of patterns of time series sequences for time series representation. Pattern representation ensures input and output data unification through trend filtering and variance equalization. This simplifies the forecasting problem and allows us to use models based on pattern similarity. We consider four such models: nearest-neighbor model, fuzzy neighborhood model, kernel regression model, and general regression neural network. Three variants of the approach were proposed. A basic one and two hybrid solutions combining similarity-based and statistical methods (ARIMA and exponential smoothing).In the experimental part of the work, the proposed models were used to forecast the monthly electricity demand in 35 European countries. The results show the high performance of the proposed models, which outperform both the comparative classical statistical models and machine learning models in terms of accuracy, simplicity, and ease of optimization. Among the proposed variants, a hybrid approach combining similarity-based methods with exponential smoothing turned out to be the most accurate. The study highlights the many advantages of the proposed pattern similarity-based models such as clear operation principles, a small number of parameters to adjust, no training procedure, fast optimization procedure, good generalization ability, ability to work on the newest data without retraining, and delivery of multi-step forecasts.

Determination of vadose zone and saturated zone nitrate lag times using long-term groundwater monitoring data and statistical machine learning

Abstract. In this study, we explored the use of statistical machine learning and long-term groundwater nitrate monitoring data to estimate vadose zone and saturated zone lag times in an irrigated alluvial agricultural setting. Unlike most previous statistical machine learning studies that sought to predict groundwater nitrate concentrations within aquifers, the focus of this study was to leverage available groundwater nitrate concentrations and other environmental variables to determine mean regional vertical velocities (transport rates) of water and solutes in the vadose zone and saturated zone (3.50 and 3.75 m yr−1, respectively). The statistical machine learning results are consistent with two primary recharge processes in this western Nebraska aquifer, namely (1) diffuse recharge from irrigation and precipitation across the landscape and (2) focused recharge from leaking irrigation conveyance canals. The vadose zone mean velocity yielded a mean recharge rate (0.46 m yr−1) consistent with previous estimates from groundwater age dating in shallow wells (0.38 m yr−1). The saturated zone mean velocity yielded a recharge rate (1.31 m yr−1) that was more consistent with focused recharge from leaky irrigation canals, as indicated by previous results of groundwater age dating in intermediate-depth wells (1.22 m yr−1). Collectively, the statistical machine learning model results are consistent with previous observations of relatively high water fluxes and short transit times for water and nitrate in the primarily oxic aquifer. Partial dependence plots from the model indicate a sharp threshold in which high groundwater nitrate concentrations are mostly associated with total travel times of 7 years or less, possibly reflecting some combination of recent management practices and a tendency for nitrate concentrations to be higher in diffuse infiltration recharge than in canal leakage water. Limitations to the machine learning approach include the non-uniqueness of different transport rate combinations when comparing model performance and highlight the need to corroborate statistical model results with a robust conceptual model and complementary information such as groundwater age.

Forecasting the dynamics of cumulative COVID-19 cases (confirmed, recovered and deaths) for top-16 countries using statistical machine learning models: Auto-Regressive Integrated Moving Average (ARIMA) and Seasonal Auto-Regressive Integrated Moving Average (SARIMA)

Most countries are reopening or considering lifting the stringent prevention policies such as lockdowns, consequently, daily coronavirus disease (COVID-19) cases (confirmed, recovered and deaths) are increasing significantly. As of July 25th, there are 16.5 million global cumulative confirmed cases, 9.4 million cumulative recovered cases and 0.65 million deaths. There is a tremendous necessity of supervising and estimating future COVID-19 cases to control the spread and help countries prepare their healthcare systems. In this study, time-series models — Auto-Regressive Integrated Moving Average (ARIMA) and Seasonal Auto-Regressive Integrated Moving Average (SARIMA) are used to forecast the epidemiological trends of the COVID-19 pandemic for top-16 countries where 70%–80% of global cumulative cases are located. Initial combinations of the model parameters were selected using the auto-ARIMA model followed by finding the optimized model parameters based on the best fit between the predictions and test data. Analytical tools Auto-Correlation function (ACF), Partial Auto-Correlation Function (PACF), Akaike Information Criterion (AIC) and Bayesian Information Criterion (BIC) were used to assess the reliability of the models. Evaluation metrics Mean Absolute Error (MAE), Mean Square Error (MSE), Root Mean Square Error (RMSE) and Mean Absolute Percent Error (MAPE) were used as criteria for selecting the best model. A case study was presented where the statistical methodology was discussed in detail for model selection and the procedure for forecasting the COVID-19 cases of the USA. Best model parameters of ARIMA and SARIMA for each country are selected manually and the optimized parameters are then used to forecast the COVID-19 cases. Forecasted trends for confirmed and recovered cases showed an exponential rise for countries such as the United States, Brazil, South Africa, Colombia, Bangladesh, India, Mexico and Pakistan. Similarly, trends for cumulative deaths showed an exponential rise for countries Brazil, South Africa, Chile, Colombia, Bangladesh, India, Mexico, Iran, Peru, and Russia. SARIMA model predictions are more realistic than that of the ARIMA model predictions confirming the existence of seasonality in COVID-19 data. The results of this study not only shed light on the future trends of the COVID-19 outbreak in top-16 countries but also guide these countries to prepare their health care policies for the ongoing pandemic. The data used in this work is obtained from publicly available John Hopkins University’s COVID-19 database.

Predicting Energy Demand in Semi-Remote Arctic Locations

Forecasting energy demand within a distribution network is essential for developing strategies to manage and optimize available energy resources and the associated infrastructure. In this study, we consider remote communities in the Arctic located at the end of the radial distribution network without alternative energy supply. Therefore, it is crucial to develop an accurate forecasting model to manage and optimize the limited energy resources available. We first compare the accuracy of several models that perform short-and medium-term load forecasts in rural areas, where a single industrial customer dominates the electricity consumption. We consider both statistical methods and machine learning models to predict energy demand. Then, we evaluate the transferability of each method to a geographical rural area different from the one considered for training. Our results indicate that statistical models achieve higher accuracy on longer forecast horizons relative to neural networks, while the machine-learning approaches perform better in predicting load at shorter time intervals. The machine learning models also exhibit good transferability, as they manage to predict well the load at new locations that were not accounted for during training. Our work will serve as a guide for selecting the appropriate prediction model and apply it to perform energy load forecasting in rural areas and in locations where historical consumption data may be limited or even not available.

Using weak supervision and deep learning to classify clinical notes for identification of current suicidal ideation

Mental health concerns, such as suicidal thoughts, are frequently documented by providers in clinical notes, as opposed to structured coded data. In this study, we evaluated weakly supervised methods for detecting “current” suicidal ideation from unstructured clinical notes in electronic health record (EHR) systems. Weakly supervised machine learning methods leverage imperfect labels for training, alleviating the burden of creating a large manually annotated dataset. After identifying a cohort of 600 patients at risk for suicidal ideation, we used a rule-based natural language processing approach (NLP) approach to label the training and validation notes (n = 17,978). Using this large corpus of clinical notes, we trained several statistical machine learning models—logistic classifier, support vector machines (SVM), Naive Bayes classifier—and one deep learning model, namely a text classification convolutional neural network (CNN), to be evaluated on a manually-reviewed test set (n = 837). The CNN model outperformed all other methods, achieving an overall accuracy of 94% and a F1-score of 0.82 on documents with “current” suicidal ideation. This algorithm correctly identified an additional 42 encounters and 9 patients indicative of suicidal ideation but missing a structured diagnosis code. When applied to a random subset of 5,000 clinical notes, the algorithm classified 0.46% (n = 23) for “current” suicidal ideation, of which 87% were truly indicative via manual review. Implementation of this approach for large-scale document screening may play an important role in point-of-care clinical information systems for targeted suicide prevention interventions and improve research on the pathways from ideation to attempt.

Integration of grey analysis with artificial neural network for classification of slope failure

With the advent of technology and the introduction of computational intelligent methods, the prediction of slope failure using the machine learning (ML) approach is rapidly growing for the past few decades. This study employs an “artificial neural network” (ANN) to predict the slope failures based on historical circular slope cases. Using the feed-forward back-propagation algorithm with a multilayer perceptron network, ANN is a powerful ML method capable of predicting the complex model of slope cases. However, the prediction result of ANN can be improved by integrating the statistical analysis method, namely grey relational analysis (GRA), to the ANN model. GRA is capable of identifying the influencing factors of the input data based on the correlation level of the reference sequence and comparability sequence of the dataset. This statistical machine learning model can analyze the slope data and eliminate the unnecessary data samples to improve the prediction performance. Grey relational analysis-artificial neural network (GRANN) prediction model was developed based on six slope factors: unit weight, friction angle, cohesion, pore pressure ratio, slope height, and slope angle, with the factor of safety (FOS) as the output factor. The prediction results were analyzed based on accuracy percentage and receiver operating characteristic (ROC) values. It shows that the GRANN model has outperformed the ANN model by giving 99% accuracy and 0.999 ROC value, compared with 91% and 0.929.

Linear Model Selection and Regularization for Serum Prostate-Specific Antigen Prediction of Patients With Prostate Cancer Using R

Prostate cancer is the commonly diagnosed cancer worldwide, and there were 1,276 thousand new prostate cancer cases and 359 thousand deaths in 2018. Prostate-specific antigen (PSA) blood level is often elevated in men with prostate cancer, so PSA testing can detect prostate tumours when they are small, low-grade, and localized. The PSA testing is hard to apply on the less developed and poor areas without sufficient medical funds, so the early accurate PSA level prediction by statistical machine learning models is significant to avoid later stages of prostate cancer that spread outside the Prostate. In this literature, we compare three linear model selection and regularization methods (shrinkage, subset selection, dimension reduction) and nine candidate models (OLS regression, Ridge regression, Lasso regression, Elastic net, best subset selection, forward subset selection, backward subset selection, PCR, PLS) based on leave-one-out-cross-validation (LOOCV) prediction error. As the selection criteria leave-one-out cross-validation is sensitive to outliers, Mahalanobis distance is used for outlier detection and deletion before running each model. The shrinkage method (only lasso and elastic net models) and subset selection method (based on adjusted <inline-formula> <tex-math notation="LaTeX">$R^{2}$ </tex-math></inline-formula>, BIC, Cp, and cross-validation prediction error) can select the variables out. The feature selection results show that prostate weight, cancer volume, amount of benign prostatic hyperplasia, and whether seminal vesicle invasion is necessary variables must include predicting PSA. Age and capsular penetration are the least important variables. The variables of Gleason score, a percent of Gleason scores 4 or 5 are essential sometimes. All the diagnostic figures and results are coded by R, open access, and published on IEEE Xplore Code Ocean.

A comparative study of supervised machine learning approaches for slope failure production

Over the years, machine learning, which is a well-known method in artificial intelligent (AI) field has become a new trend and extensively applied in various applications to solve a realworld problem. This includes slope failure prediction. Slope failure is among the major geo-hazard phenomenon which gives the significant impact to the environment or human beings. The estimation of slope failure in slope stability analysis is a complex geotechnical engineering problem that involves many factors such as geology, topography, atmosphere, and land occupancy. Generally, slope failure can be estimated based on traditional methods such as limit equilibrium method (LEM) or finite equilibrium method (FEM). However, beside the methods are quite tedious and time consuming, LEM and FEM have their own limitations and do not guarantee the effectiveness when dealing against problem with various geometry or assumptions. Hence, the introduction of machine learning approaches provides the alternative tools for the prediction of slope failure. Current study applies two mostly used supervised machine learning approaches, support vector machine (SVM) and decision tree (DT) to predict the slope failure based on classification problem using historical cases. 148 of slope cases with six input parameters namely “unit weight, cohesion, internal friction angle, slope angle, slope height and pore pressure ratio and factor of safety (FOS) as an output parameter”, was collected from multinational dataset that has been extracted from the literature. For development of the prediction model, the slope data was divided into 80% training data and 20% testing data. The prediction result from testing data was validated based on statistical analysis. The result shows that SVM model has outperformed DT model by giving the prediction accuracy of 97%. ith the advent of technology and the introduction of computational intelligent methods, the prediction of slope failure using the machine learning (ML) approach is rapidly growing for the past few decades. This study employs an “artificial neural network” (ANN) to predict the slope failures based on historical circular slope cases. Using the feed-forward backpropagation algorithm with a multilayer perceptron network, ANN is a powerful ML method capable of predicting the complex model of slope cases. However, the prediction result of ANN can be improved by integrating the statistical analysis method, namely grey relational analysis (GRA), to the ANN model. GRA is capable of identifying the influencing factors of the input data based on the correlation level of the reference sequence and comparability sequence of the dataset. This statistical machine learning model can analyze the slope data and eliminate the unnecessary data samples to improve the prediction performance. Grey relational analysis-artificial neural network (GRANN) prediction model was developed based on six slope factors: unit weight, friction angle, cohesion, pore pressure ratio, slope height, and slope angle, with the factor of safety (FOS) as the output factor. The prediction results were analyzed based on accuracy percentage and receiver operating characteristic (ROC) values. It shows that the GRANN model has outperformed the ANN model by giving 99% accuracy and 0.999 ROC value, compared with 91% and 0.929.

Evaluating Deep and Statistical Machine Learning Models in the Classification of Breast Cancer from Digital Mammograms

The application of artificial intelligence techniques in computer aided detection and diagnosis problems has been among the most promising areas with interest from the scientific community and healthcare industry. Recently, deep learning has become the prime tool for such application with many studies focusing on developing variants that optimize diagnostic performance. Despite the widely accepted success of this class of techniques in this application by the scientific community, it is not prudent to consider it as the only tool available for such purpose. In particular, statistical machine learning offers a variety of techniques that can also be applied at a much lower computational cost. Unfortunately, the results from both strategies cannot be directly compared due to the differences in experimental setups and datasets used in available research studies. Therefore, we focus in this study on this direct comparison using the same dataset and similar data preprocessing as the input to both. We compare statistical machine learning to deep learning in the context of computer-aided detection of breast cancer from mammographic images. The results are compared using diagnostic performance metrics and suggest that simpler statistical machine learning techniques may provide better performance with simpler architectures that allow explanation of results.

Making Deep Learning-Based Predictions for Credit Scoring Explainable

Credit scoring has become an important risk management tool for money lending institutions. Over the years, statistical and classical machine learning models have been the most researched risk management tools in credit scoring literature, and recently the focus has turned to deep learning models. This transition is due to better performances that are shown by deep learning models in different domains. Despite deep learning models' superior performances, there is still a need for explaining how these models make their predictions. The non-transparency nature of deep learning models has created a bottleneck for their use in credit scoring. Explanations of decisions are important for lending institutions since it is a requirement for automated decisions that are generated by non-transparent models to be explained. The other issue in using deep learning models, specifically 2D Convolutional Neural Networks (CNNs), in credit scoring is the need to have the data in image format. We propose an explainable deep learning model for credit scoring which can harness the performance benefits offered by deep learning and yet comply with the legislation requirements for the automated decision-making processes. The proposed method converts tabular datasets into images and thus allowing the application of 2D CNNs in credit scoring. Each pixel of the image corresponds to a feature bin of the tabular dataset. The predictions from the 2D CNNs were explained using state-of-the-art explanation methods. Furthermore, explanations were evaluated using a sanity check methodology and also performances of the explanation methods were compared quantitatively. The proposed explainable deep learning model outperforms the other credit scoring methods on publicly available credit scoring datasets.

Context and Machine Learning Based Trust Management Framework for Internet of Vehicles

Trust is one of the core components of any ad hoc network security system. Trust management (TM) has always been a challenging issue in a vehicular network. One such developing network is the Internet of vehicles (IoV), which is expected to be an essential part of smart cities. IoV originated from the merger of Vehicular ad hoc networks (VANET) and the Internet of things (IoT). Security is one of the main barriers in the on-road IoV implementation. Existing security standards are insufficient to meet the extremely dynamic and rapidly changing IoV requirements. Trust plays a vital role in ensuring security, especially during vehicle to vehicle communication. Vehicular networks, having a unique nature among other wireless ad hoc networks, require dedicated efforts to develop trust protocols. Current TM schemes are inflexible and static. Predefined scenarios and limited parameters are the basis for existing TMmodels that are not suitable for vehicle networks. The vehicular network requires agile and adaptive solutions to ensure security, especially when it comes to critical messages. The vehicle network's wireless nature increases its attack surface and exposes the network to numerous security threats. Moreover, internet involvement makes it more vulnerable to cyberattacks. The proposedTMframework is based on context-based cognition and machine learning to be best suited to IoV dynamics. Machine learning is the best solution to utilize the big data produced by vehicle sensors. To handle the uncertainty Bayesian machine learning statistical model is used. The proposed framework can adapt scenarios dynamically and infer using the maximum possible parameter available. The results indicated better performance than existing TM methods. Furthermore, for future work, a high-level machine learning model is proposed. © 2021 Tech Science Press. All rights reserved.

A zero altered Poisson random forest model for genomic-enabled prediction.

In genomic selection choosing the statistical machine learning model is of paramount importance. In this paper, we present an application of a zero altered random forest model with two versions (ZAP_RF and ZAPC_RF) to deal with excess zeros in count response variables. The proposed model was compared with the conventional random forest (RF) model and with the conventional Generalized Poisson Ridge regression (GPR) using two real datasets, and we found that, in terms of prediction performance, the proposed zero inflated random forest model outperformed the conventional RF and GPR models.

A Systematic Review of Statistical and Machine Learning Methods for Electrical Power Forecasting with Reported MAPE Score

Electric power forecasting plays a substantial role in the administration and balance of current power systems. For this reason, accurate predictions of service demands are needed to develop better programming for the generation and distribution of power and to reduce the risk of vulnerabilities in the integration of an electric power system. For the purposes of the current study, a systematic literature review was applied to identify the type of model that has the highest propensity to show precision in the context of electric power forecasting. The state-of-the-art model in accurate electric power forecasting was determined from the results reported in 257 accuracy tests from five geographic regions. Two classes of forecasting models were compared: classical statistical or mathematical (MSC) and machine learning (ML) models. Furthermore, the use of hybrid models that have made significant contributions to electric power forecasting is identified, and a case of study is applied to demonstrate its good performance when compared with traditional models. Among our main findings, we conclude that forecasting errors are minimized by reducing the time horizon, that ML models that consider various sources of exogenous variability tend to have better forecast accuracy, and finally, that the accuracy of the forecasting models has significantly increased over the last five years.

Comparison of machine learning methods for predicting viral failure: a case study using electronic health record data.

Human immunodeficiency virus (HIV) viral failure occurs when antiretroviral therapy fails to suppress and sustain a person's viral load count below 1,000 copies of viral ribonucleic acid per milliliter. For those newly diagnosed with HIV and living in a setting where healthcare resources are limited, such as a low- and middle-income country, the World Health Organization recommends viral load monitoring six months after initiation of antiretroviral treatment and yearly thereafter. Deviations from this schedule are made in cases where viral failure occurs or at the discretion of the clinician. Failure to detect viral failure in a timely fashion can lead to delayed administration of essential interventions. Clinical prediction models based on information available in the patient medical record are increasingly being developed and deployed for decision support in clinical medicine and public health. This raises the possibility that prediction models can be used to detect potential for viral failure in advance of viral measurements, particularly when those measurements occur infrequently. Our goal is to use electronic health record data from a large HIV care program in Kenya to characterize and compare the predictive accuracy of several statistical machine learning methods for predicting viral failure at the first and second measurements following initiation of antiretroviral therapy. Predictive accuracy is measured in terms of sensitivity, specificity and area under the receiver-operator characteristic curve. We trained and cross-validated 10 statistical machine learning models and algorithms on data from over 10,000 patients in the Academic Model Providing Access to Healthcare care program in western Kenya. These included parametric, non-parametric, ensemble, and Bayesian methods. The input variables included 50 items from the clinical record, hand picked in consultation with clinician experts. Predictive accuracy measures were calculated using 10-fold cross validation. Viral load failure rate is about 20% in this patient cohort at both the first and second measurements. Ensemble techniques generally outperformed other methods. For predicting viral failure at the first follow up measure, specificity was over 90% for these methods, but sensitivity was typically in the 50-60% range. Predictive accuracy was greater for the second follow up measure, with sensitivities over 80%. Super Learner, gradient boosting and Bayesian additive regression trees consistently outperformed other methods. For a viral failure rate of 20%, the positive predictive value for the top-performing methods is between 75 and 85%, while the negative predictive value is over 95%. Evidence from this study suggests that machine learning techniques have potential to identify patients at risk for viral failure prior to their scheduled measurements. Ultimately, prognostic virologic assessment can help guide the administration of earlier targeted intervention such as enhanced drug resistance monitoring, rigorous adherence counseling, or appropriate next-line therapy switching. External validation studies should be used to confirm the results found here.