Detecting new obfuscated malware variants: A lightweight and interpretable machine learning approach

Predicting depression using serum perfluoroalkyl and polyfluoroalkyl substances levels via interpretable machine learning.

Pharmaceutical contamination of aquatic systems poses an increasing environmental concern due to the persistence, bioactivity, and incomplete removal of these compounds by conventional wastewater treatment processes. This study investigates the adsorptive removal of three structurally distinct pharmaceuticals: ketotifen fumarate (KF), doxycycline hyclate (DXC), and nystatin (Nyst), using raw bentonite (RB). By combining batch experiments with an interpretable machine learning (ML) framework, adsorption kinetics, equilibrium, and thermodynamics were evaluated. Additionally, four Ant Lion Optimizer (ALO)-optimized models: Artificial Neural Network (ANN), Support Vector Regression (SVR), Random Forest (RF), and eXtreme Gradient Boosting (XGBoost), were employed to predict adsorption capacity under diverse conditions. RB exhibited high adsorption capacities: 178.86 ± 1.26 mg/g for KF, 222.91 ± 2.02 mg/g for DXC, and 190.25 ± 2.86 mg/g for Nyst. Adsorption equilibrium was best described by the Freundlich isotherm, indicating multilayer adsorption on a heterogeneous surface. Thermodynamic and spectroscopic analyses revealed a dual mechanism involving electrostatic attraction, hydrogen bonding, cation exchange, and van der Waals interactions, with DXC and Nyst adsorption being endothermic and KF adsorption exothermic. SHAP (SHapley Additive exPlanations) analysis identified adsorbent dosage, initial concentration, and pH as dominant operational factors, while the molecular descriptor nC (number of carbon atoms) emerged as key to differentiating pharmaceuticals, linking larger molecular size to stronger adsorption. The XGBoost model achieved the highest accuracy (R2 = 0.972, RMSE = 0.1225), demonstrating robust generalizability. These findings highlight RB as a low-cost, scalable adsorbent and establish an interpretable ML approach capable of linking molecular structure to adsorption behavior.

This study aimed to develop and validate an interpretable machine learning (ML) model based on nnU-Net automated segmentation and computed tomography (CT) radiomics for preoperative risk stratification in thymic epithelial tumors (TETs). In this retrospective multicenter study, 764 patients with pathologically confirmed TETs were enrolled and divided into training, internal validation, and two external validation cohorts. An nnU-Net model was trained for automatic tumor segmentation, with performance assessed by the dice similarity coefficient (DSC). Radiomic features were extracted from the automated segmentations of venous-phase CT images, and least absolute shrinkage and selection operator (LASSO) regression was applied for feature selection. Predictive models, including radiomics-only, clinical-only, and a clinical-radiomics (combined) model, were constructed using five ML algorithms (RF, SVM, KNN, DT, and LR). Model performance was evaluated using the receiver operating characteristic (ROC) curve. Delong’s test was employed to compare these ML models and select the best-performing model as the final model. Calibration curve and decision curve analysis (DCA) were performed to assess clinical efficacy of the final model. The interpretability of the optimal model was elucidated using SHapley Additive exPlanations (SHAP). The nnU-Net segmentation model achieved excellent performance, with a DSC of 0.979 on the test cohort. Compared to the other four combined models, the RF-based combined model demonstrated superior predictive efficacy, yielding area under the curve (AUC) values of 0.941 (training), 0.884 (internal validation), 0.867 (external validation 1), and 0.872 (external validation 2). The calibration curves indicated excellent agreement between the RF-based model’s predictions and actual outcomes, and furthermore, DCA confirmed its superior net benefit over baseline strategies across a wide range of thresholds. SHAP tool identified 11 radiomic features and 3 clinical features as the most influential features, providing transparency into the model’s decision-making process. The nnU-Net framework enables accurate and efficient automatic segmentation of TETs. The proposed RF-based combined model, integrating clinical and radiomic features, provides a robust and interpretable tool for identifying the high-risk TETs, holding promise for supporting clinical decision-making towards personalized therapy.

High‐throughput computational screening and machine learning hold significant potential for exploring diverse chemical compositions and discovering novel inorganic solids. However, the complexity of point defects, which occur in all inorganic solids and are often crucial to their functionality and synthesizability, presents significant challenges. Here, this study presents a defect‐chemistry‐trained, interpretable machine learning approach, designed to accelerate the exploration and discovery of unconventional proton‐conducting inorganic solid electrolytes. By considering dopant dissolution and hydration reactions, the machine learning models provide quantitative predictions and physical interpretations for synthesizable host–dopant combinations with hydration capabilities across various structures. Utilizing these insights, two unconventional proton conductors, Pb‐doped Bi12SiO20 sillenite and eulytite‐type Sr‐doped Bi4Ge3O12, are discovered in the first two synthesis trials. Notably, the Pb‐doped Bi12SiO20 represents an unprecedented class of proton‐conducting electrolyte composed solely of groups 14 and 15 cations and featuring a sillenite structure. It exhibits unique and fast 3D proton conduction along a loosely bonded BiO5 network. This study demonstrates an efficient approach for exploring novel inorganic materials.

Coastal shorelines are dynamic by nature, evolving in reaction to hydro-geomorphic processes across the coast. Coastal shoreline change is accelerating globally due to shifts in land use brought on by coastal urbanisation and increasing human population pressures. In order to develop suitable risk management alternatives and ensure long-term management of populations, infrastructure, and ecosystems, coastline location over time and coastal erosion patterns are crucial for addressing current and future climate change scenarios. However, achieving this purpose is particularly challenging on sandy shores that slope gently, where even slight variations in sea level cause notable morphological changes. The coastal areas of India are both physiologically productive and highly populated. However, they are susceptible to erosion and deposition from both natural disasters and human activity. When assessing shoreline dynamics, these threats have been given priority as part of the sustainable management of coastal zones. In this study, we show how the artificial neural network (ANN), and support vector machine (SVM) algorithms can be used as interpretable machine learning (ML) models to measure changes in the shoreline. The ML model contains weights that are multiplied with relevant inputs/features to obtain a better prediction. This study showed how well Earth Observation and geographic information systems may be combined to provide comprehensive, long-term research on coastal change. According to the data, the rate of erosion along the Chennai coast varies between -0.2 and -2.5 m/year. Along the Chennai coast, accretion rates vary from 1 to 4.6 meters per year. With erosion rates ranging from -0.1 to -6.8 m/y and accretion rates ranging from 0.2 to 5.0 m/y, the shoreline of Vishakhapatnam displays a regular pattern of both processes. The accretion rate along the Puri coast varies between 0.1 and 3.22 m/y. Given how crucial these coastal towns are to India's cultural and economic endeavours, the alterations to the shorelines of these three metropolises are quite concerning. In addition to raising sea levels globally, climate change and global warming are intensifying and increasing the frequency of extreme occurrences like tropical cyclones in the Bay of Bengal, which includes these three shores. The coasts of these urban areas may shift due to a range of human activities and natural events like tropical storms and rising sea levels. For the coastal regions of Vishakhapatnam, Puri, Chennai, and other Indian coastal towns with comparable physical attributes, this study may aid in the development of suitable management plans and regulations.

BackgroundCardiovascular-Kidney-Metabolic (CKM) syndrome is a systemic disease characterized by pathophysiological interactions between the cardiovascular system, chronic kidney disease, and metabolic risk factors. In China, the prevalence of CKM in middle-aged and elderly patients is relatively high. The current research lacks an exploration into the impact of social and environmental determinants of health on depression in CKM patients.ObjectiveThis study aims to construct a depression risk prediction model for middle-aged and elderly CKM patients by social and environmental determinants of health.MethodsIn this study, 3220 participants were included and collected from three waves of the China Health and Retirement Longitudinal Study (CHARLS). A depression risk prediction model for middle-aged and elderly CKM patients was constructed by using 10 machine learning models. Additionally, the mediating effect of NO2 between arthritis and depression outcomes was analyzed in this population.ResultsAn interpretable machine learning model framework was constructed to predict depression risk in middle-aged and elderly CKM patients using the longitudinal cohort data from CHARLS. The RF model demonstrated strong performance in predicting the training set, and the Xgboost model exhibited excellent generalization ability. The presence of arthritis showed a significant independent effect on depression outcomes, with an average direct effect of − 8.5559. The total effect of arthritis on depression outcomes was − 9.5162. The mediating effect of NO2 represented 10.09% of the total effect (average), indicating that NO2 serves as a mediator between arthritis and depression outcomes.ConclusionsA depression risk prediction model for middle-aged and elderly CKM patients was developed based on the CHARLS longitudinal data from 2011 to 2015. The SHAP framework was used to provide machine learning model explanations. Intervention strategies that address social and environmental determinants of health are needed. Potential strategies include enhancing urban greening to reduce NO2 levels, integrating CKM as a special outpatient chronic disease to alleviate the financial burdens of patients, and focusing on the treatment of arthritis and digestive diseases in CKM patients.

Interpretable machine learning approach in estimating traffic volume on low-volume roadways

According to the WHO (World Health Organization), infectious diseases continue to one of the leading causes of death worldwide. Since the core microbiota flora of humans is largely diverse and HGT (Horizontal Gene Transfer), it is very challenging to determine whether a particular bacterial strain is commensal or pathogenic to humans. With the latest advances in NGS (Next-Generation Sequencing) technology, bioinformatics tools and techniques using NGS data have increasingly been used for the diagnosis and monitoring of infectious diseases. Even if the biological background is not available, the machine learning method can still infer the pathogenic phenotype from the NGS readings, independent of the database of known organisms, and being studied intensively. However, previous methods have not considered opportunistic pathogenic and interpretability of black box model, are not well suited for clinical requirements.

Staphylococcus aureus bloodstream infection (SA-BSI) is a life-threatening condition in ICU patients, often leading to progressive multi-organ dysfunction. Traditional static assessments may underestimate the dynamic nature of organ failure. We aimed to identify distinct organ dysfunction trajectories and evaluate their prognostic significance using a data-driven and interpretable machine learning approach. ICU patients with SA-BSI from two independent cohorts admitted between 2008 and 2024 were retrospectively analyzed (MIMIC-IV, n=834; the Second Affiliated Hospital of Zhejiang University School of Medicine (SAHZU), n=151). Daily Sequential Organ Failure Assessment (SOFA) scores from Day -1 to +3 were used to derive trajectory subgroups via group-based trajectory modeling. Associations with in-hospital mortality were assessed using multivariable Cox regression and Kaplan-Meier analysis. An XGBoost model was developed to predict trajectory group membership based on baseline ICU admission variables, with interpretability assessed via SHAP values. Three reproducible SOFA trajectory groups were identified in both cohorts, representing stable, moderately worsening, and severely deteriorating clinical courses. Compared with the stable group, patients in the severely deteriorating group had a markedly increased risk of in-hospital mortality (HR 4.60, 95% CI 3.49-6.07), with consistent effects observed across both cohorts. The XGBoost model demonstrated strong predictive performance for identifying severely deteriorating trajectories (AUC 0.96), and SHAP analysis revealed biologically coherent predictors underlying each trajectory. Early ICU data can predict dynamic organ dysfunction trajectories in SA-BSI patients. Trajectory-based phenotyping, combined with interpretable machine learning, offers a clinically valuable framework for early risk stratification and individualized ICU management.

Auroral electrojet (AE) activity is a widely used system-level indicator of how the magnetosphere-ionosphere system responds to solar wind driving across a wide range of spatial and temporal scales. While large-scale solar wind parameters are known to control overall levels of geomagnetic activity, auroral responses often show substantial variability under similar upstream conditions. This suggests that additional aspects of solar wind variability, beyond mean magnetic field and plasma properties, may influence how energy and momentum are transferred into the coupled system. In particular, the role of multiscale solar wind turbulence and structured variability in modulating auroral activity remains incompletely understood.In this work, we examine how multiscale solar wind turbulence contributes to auroral variability using an interpretable machine learning approach. OMNI observations are combined with AE index measurements to construct models that integrate conventional bulk drivers with measures describing solar wind variability across multiple timescales. Interpretable diagnostics are then used to assess how turbulence-related information influences auroral responses under different upstream conditions. While the overall improvement in forecasting skill obtained by including turbulence measures is modest, the results reveal consistent and scale-dependent contributions associated with structured solar wind variability. These findings suggest that solar wind turbulence plays a secondary but informative role in shaping auroral activity, providing insight into how mesoscale variability can modulate system-level coupling and highlighting the value of interpretable machine learning for advancing both physical understanding and space weather prediction.

Zenith Total Delay (ZTD) derived from Global Navigation Satellite System (GNSS) observations is a critical parameter for precise positioning and atmospheric studies. ZTD is continuously estimated at GNSS stations, forming high-resolution temporal time series that reflect the dynamic behavior of the troposphere. In recent years, deep learning approaches have been increasingly applied to ZTD estimation due to their ability to model nonlinear relationships. However, particularly for time-series reconstruction problems, it remains an open question whether the added architectural complexity of such models is always necessary.In this study, feature-driven classical machine learning models and a data-driven neural network approach are systematically compared for reconstructing missing ZTD values in GNSS time series. The analysis is based on ZTD observations from six International GNSS Service (IGS) stations covering the period from February 2023 to January 2024. All models are trained using an identical feature set comprising lagged ZTD values, tropospheric gradients, ZTD variances, station coordinates, and temporal attributes. This design ensures a fair and interpretable comparison between different modeling approaches.Linear Regression is considered as a baseline model, while Random Forest represents a nonlinear yet interpretable machine learning approach, and a Fully Connected Neural Network (FNN) is employed as a deep learning model. Model performance is evaluated using Mean Absolute Error (MAE), Root Mean Square Error (RMSE), and the coefficient of determination (R²), with a leave-one-station-out validation strategy applied to assess generalization capability.The results indicate that the Random Forest model achieves accuracy comparable to that of the FNN, while exhibiting greater stability and consistency across stations. The results highlight that incorporating physically meaningful features into the input space can be as effective as increasing model complexity for ZTD reconstruction. The study provides methodological and practical insights for selecting appropriate modeling strategies in tropospheric delay estimation.

Machine Learning (ML) models are powerful tools for meteorological applications but often operate as "black boxes", hindering scientific understanding. This study addresses this challenge by implementing an interpretable ML approach Rulefit for Quantitative Precipitation Estimation (QPE). The algorithm was applied to a multisource dataset from eastern China, incorporating composite radar reflectivity from China’s New Generation DopplerWeather Radar, Digital Elevation Model data, and Himawari-8 satellite bands 7-10. A geographical mask based on China’s national borders was applied to exclude data points outside the land area. Additionally, standard scaling normalization was applied to all input features to account for different units and value ranges across datasets. The model’s performance was evaluated against six baseline models and traditional Z-R relationship. The RuleFit model demonstrated strong predictive performance, achieving a Critical Success Index (CSI) of 0.6015 and a Probability of Detection (POD) of 0.9359 for 10-minute precipitation events exceeding a 2mm threshold. This accuracy was comparable to other ML models and significantly surpassed traditional Z-R relationship methods. Crucially, the generated decision rules and Partial Dependence Plots(PDP) provided transparent insights into the model’s logic, revealing key non-linear interactions between radar and satellite features and showing an emphasis on predicting heavy rainfall. Our findings show that RuleFit is not only an accurate QPE tool but also a framework for uncovering meteorological relationships, thereby building researcher confidence and addressing the critical need for interpretability in ML-based atmospheric science.

The continuously increasing demand for electricity presents significant issues for policymakers and utility companies. An accurate forecast of electricity usage is critical for effective energy management. This thesis provides an interpretable machine-learning approach for forecasting energy demand that incorporates macroeconomic variables such as GDP, inflation, and industrial growth. The study uses data from the World Development Indicators from 1973 to 2021 and employs the multi-model. Our findings emphasize the importance of economic considerations on electricity demand, resulting in a reliable method for forecasting energy consumption. This study adds to the body of knowledge by providing a clear model to aid in decision-making processes connected to energy management and policy creation.

Background Little is known about acute kidney injury (AKI) and acute kidney disease (AKD) in patients with chronic obstructive pulmonary disease (COPD) and COPD mortality based on the acute/subacute renal injury. This study develops machine learning models to predict AKI, AKD, and mortality in COPD patients, utilizing web applications for clinical decisions. Methods We included 2,829 inpatients from January 2016 to December 2018. Data were split into 80% for training and 20% for testing. Eight machine learning algorithms were used, and model performance was evaluated using various metrics. SHAP was used to visualize the decision process. The best models, assessed using AUROC were used to develop web applications for identifying high-risk patients. Results The incidence rates were 13.71% for AKI and 15.11% for AKD. The overall mortality rate was 4.84%. LightGBM performed best with AUROC of 0.815, 0.827, and 0.934 in AKI, AKD, and mortality, respectively. Key predictors for AKI were Scr, neutrophil percentage, cystatin c, BUN, and LDH. For AKD, the key predictors were age, AKI grade, HDL-C, Scr, and BUN. The key predictors for mortality included the use of dopamine and epinephrine drugs, cystatin c, renal function trajectory, albumin, and neutrophil percentage. Force plots visualized the prediction process for individual patients. Conclusions The incidence of AKI and AKD is significant in patients with COPD. Renal function trajectory is crucial for predicting mortality in these patients. Web applications were developed to predict AKI, AKD, and mortality, improving prognosis by identifying high-risk patients and reducing adverse events and disease progression.

This research tries to establish a communication bridge between the decision making of a practicing physician and a machine learning model. Machine learning has long been used to assist in diagnosing patients through deep learning and medical image analysis. However, in medical science, it is possible to have multiple right answers when treating patients. This leaves the practitioners room to exercise their judgement on what would be best for the patient. The machine learning model can be a communication tool to assist the practitioners in reach the final decision. In this paper, the machine learning model used for this experiment is XGBoost, a gradient boosted decision tree. Through the experiment with the real patient data, we show how decision tree constructions are easily readable for medical determination.