Interpretable Machine Learning Models Research Articles

Optimized Collective Variable for Collapse Transition in Linear Hydrophobic Polymers: Importance of Hydration Water and End-to-End Distance.

Choosing an appropriate collective variable (CV) for any biomolecular process is a challenging task. Researchers are developing methods to solve this issue using a variety of methodologies, most recently using machine learning (ML) methods. In this work, we investigate the mechanism of collapse transition across various lengths of polymer systems through adaptively sampled multiple short trajectories utilizing the Time Lagged Independent Component Analysis (TICA) framework. From TICA analysis, it is revealed that the radius of gyration (Rg) and end-to-end distance serve as good order parameters (OPs) for these systems describing overall energy landscapes. Markov state model (MSM) and mean first passage time (MFPT) analysis suggest that hydration water (Nw) plays a determining role in dictating the time scale and barrier for the collapsed transition for the C40 system. P-fold analysis on identifying transition state ensembles (TSE) identified by committor analysis also strengthens the role of Nw in such a transition. TICA, MSM, and committor analyses on the collapse transition for C45 reveal similarities with C40 systems in different aspects. Furthermore, we propose a pipeline integrating XGBoost regression along with an interpretable ML model, Shapley Additive exPlanation (SHAP) to precisely elucidate the contribution of each OP locally at the TSE. Through this approach, we observe that the collapse transition is primarily driven by Nw for both polymer systems. A carefully designed protocol for the collapsed transition of C60 systems indirectly reiterates the above result. Overall, our results suggest that while the end-to-end distance should be considered for better resolution of metastable states in the landscape, Nw is the crucial coordinate to be used in enhanced sampling for the exploration of actual collapse transitions for linear hydrophobic polymer systems. The Python code for analyzing the contribution of different OPs in the TSE using an ML-aided protocol is available on GitHub (https://github.com/saikat-ai/linear_polymer_project).

Linking industrial emissions and dietary exposure to human burdens of polychlorinated naphthalenes

Relationships between toxic pollutant emissions during industrial processes and toxic pollutant dietary intakes and adverse health burdens have not yet been quantitatively clarified. Polychlorinated naphthalenes (PCNs) are typical industrial pollutants that are carcinogenic and of increasing concern. In this study, we established an interpretable machine learning model for quantifying the contributions of industrial emissions and dietary intakes of PCNs to health effects. We used the SHapley Additive exPlanations model to achieve individualized interpretability, enabling us to evaluate the specific contributions of individual feature values towards PCNs concentration levels. A strong relationship between PCN dietary intake and body burden was found using a robust large-scale PCN diet survey database for China containing the results of the analyses of 17,280 dietary samples and 4480 breast milk samples. Industrial emissions and dietary intake contributed 12 % and 52 %, respectively, of the PCN burden in breast milk. The model quantified the contributions of food consumption and industrial emissions to PCN exposure, which will be useful for performing accurate health risk assessments and developing reduction strategies of PCNs.

Interpretable machine learning models based on shear-wave elastography radiomics for predicting cardiovascular disease in diabetic kidney disease patients.

The risk of cardiovascular complications is significantly elevated in patients with diabetic kidney disease (DKD). Recognizing the link between the progression of DKD and an increased risk of cardiovascular disease (CVD), it is crucial to focus on the early prediction and management of CVD risk factors among these patients to potentially enhance their health outcomes. This study sought to bridge the existing gap by developing and validating machine learning (ML) models that utilize clinical data and shear wave elastography (SWE) radiomics features to identify patients at risk of CVD, ultimately aiming to improve the management of DKD. This study conducted a retrospective analysis of 586 patients with DKD, dividing them into training and external validation cohorts. We categorized patients based on the presence or absence of CVD. Utilizing SWE imaging, we extracted and standardized radiomics features to develop multiple ML models. These models underwent internal validation using radiomics features alone, clinical data, or a combination thereof. The optimal model was then identified, and its feature importance was assessed through the Shapley Additive Explanations (SHAP) method, before proceeding to external validation. Among the 586 patients analyzed, 30.7% (180/586) were identified as at risk for CVD. The study pinpointed six significant radiomics features related to CVD, alongside six critical pieces of clinical data. The Support Vector Machine (SVM) model outperformed others in both internal and external validations. Further, SHAP analysis highlighted five principal determinants of CVD risk, comprising three clinical indicators and two SWE radiomics features. This study highlights the effectiveness of an SVM model that combines clinical and radiomics features in predicting CVD risk among DKD patients. It enables early prediction of CVD in this patient group, thereby supporting the implementation of timely and suitable interventions.

Predicting Heat Capacity of Molecular Fluids Using Interpretable Machine Learning Model

Predicting Heat Capacity of Molecular Fluids Using Interpretable Machine Learning Model

Molecular simulations and deep neural networks‐based interpretable machine learning modelling of reverse adsorptive MOFs for ethane/ethylene separation

AbstractThe thermal decomposition of ethane (C2H6) and the steam cracking of fossil fuels are the main sources of ethylene (C2H4). However, it usually contains 5%–9% of C2H6 residue, which must be reduced to ensure its utilization during polymerization. C2H6 and C2H4 have comparable kinetic diameters and boiling points (C2H6: 4.44, 184.55 K; C2H4: 4.16, 169.42 K), which makes the separation process very difficult. This contribution employs a methodology that integrates machine learning (ML) with Monte Carlo simulations to evaluate the ddmof database to develop a predictive model for separating ethane (C2H6) and ethylene (C2H4). The ML model's input is the metal–organic frameworks (MOFs) chemical and structural descriptors. The grand canonical Monte Carlo (GCMC) simulations in RASPA software were carried out to calculate the equilibrium adsorption of ethane and ethylene. Different ML models such as random forest, decision tree, and deep neural network models have been tested to estimate the selectivity and ethane uptake from the MOF data being generated. Interpretable ML model using SHapley Additive exPlanations (SHAP) is developed for the better understanding of the impact of the parameters on selectivity and ethane uptake. A user‐friendly graphical user interface (GUI) is presented, allowing users to predict the ethane uptake and selectivity of MOFs simply by entering the values of chemical and structural descriptors.

Optimized green infrastructure planning at the city scale based on an interpretable machine learning model and multi-objective optimization algorithm: A case study of central Beijing, China

Green infrastructure (GI) has developed as a sustainable approach to the mitigation of urban floods. While machine learning (ML) models have exhibited advantages in urban flood simulation, their direct application to support the quantitative planning of GI at the city scale remains a challenge. To address this, an interpretable ML model based on support vector machine (SVM) and the Shapley additive explanations (SHAP) approach is integrated with the non-dominated sorting genetic algorithm-II (NSGA-II) in this study. The model is applied to the case of central Beijing, China, and demonstrates a robust performance with a high area under curve (AUC) value of 0.94. The results of the urban flood susceptibility assessment identify the urban-rural transition zone in the study area as being under a greater flood threat. Via model interpretation with SHAP, the dominant roles of GI and grey infrastructure (GrI) in preventing flood are revealed and the non-linear complementarity between the two is demonstrated to be more significant in study units with a GI proportion of less than 0.45. Supported by the NSGA-II-based optimization framework, optimal GI plans under different total implementations of GI are achieved, among which a solution with a 3.21% increase in the total GI area is selected as that with the best investment efficiency. The pattern of GI implementation is suggested to be dispersed and small-scale by model. This study provides a tool with broad application prospects, effectively integrating GI implementation with urban planning. The findings of this study not only provide important references for the determination of the priority areas of new ecological space in Beijing, but also provide areas that share similar characteristics with new insight into GI planning and the management of urban floods.

Exploring the response and prediction of phytoplankton to environmental factors in eutrophic marine areas using interpretable machine learning methods

Coastal marine areas are frequently affected by human activities and face ecological and environmental threats, such as algal blooms and climate change. The community structure of phytoplankton—primary producers in marine ecosystems—is highly sensitive to environmental factors, such as temperature, salinity, and nutrients. However, traditional methods for exploring the relationship between phytoplankton communities and environmental factors in eutrophic marine areas are limited by various factors. Therefore, this study employed interpretable machine learning models, integrating high-dimensional data analysis and complex system modeling, to quantitatively and thoroughly analyze the dynamic relationship between phytoplankton communities and environmental variables in high-frequency samples collected over 53 weeks from eutrophic marine areas. The cell abundance of phytoplankton exhibited a distinct “two-peak pattern” variation. Interpretable machine learning model analysis revealed the dynamic contributions of different environmental factors during changes in the phytoplankton community structure. The results showed that temperature was a key environmental factor that affected phytoplankton growth during peak periods. In addition, the contribution of salinity increased during the second peak in phytoplankton abundance, highlighting its central role in the ecological dynamics of this phase. During green tide outbreaks, particularly in Area 01, the contributions of factors such as temperature and salinity increased, whereas those of phosphates and silicates decreased, indicating that green tide outbreaks substantially altered the nutritional dynamics of the ecosystem. Furthermore, different phytoplankton species, such as Skeletonema costatum, Thalassiosira spp., and Nitzschia spp., exhibit varying responses to environmental factors. Hence, the predictions made using random forest and generalized additive models for phytoplankton cell abundance in two marine areas revealed complex nonlinear relationships between environmental factors, such as temperature, salinity, and phytoplankton abundance.

Consumer Segmentation and Decision: Explainable Machine Learning Insights

ABSTRACT The e-commerce market is experiencing a period of rapid growth, with online shopping becoming a prevalent consumer behavior. However, the challenge of identifying and segmenting high-value customer groups, and subsequently enhancing the conversion rate of purchases made on the shop’s platform, has remained a significant obstacle for merchants engaged in e-commerce. This study presents a novel approach that combines consumer involvement theory, quality signaling theory and consumer heterogeneity theory to develop a multi-algorithmic, interpretable machine learning model based on shapely additive explanations (SHAP). The results revealed the existence of four distinct consumer groups: the comprehensively involved, interactive, reading-oriented, and low-involvement groups. Comprehensively involved and interactive consumers have the highest purchase conversion rate and should be given priority attention. This study addresses the limitations of single theoretical perspective and “black box” problem of machine learning models for decision-making behavior, and can bring management insights for merchants to improve shop operation.

Interpretable machine learning model for early prediction of acute kidney injury in patients with rhabdomyolysis

Abstract Background Rhabdomyolysis (RM) is a complex set of clinical syndromes. RM-induced acute kidney injury (AKI) is a common illness in war and military operations. This study aimed to develop an interpretable and generalizable model for early AKI prediction in patients with RM. Methods Retrospective analyses were performed on 2 electronic medical record databases: the eICU Collaborative Research Database and the Medical Information Mart for Intensive Care III database. Data were extracted from the first 24 hours after patient admission. Data from the two datasets were merged for further analysis. The extreme gradient boosting (XGBoost) model with the Shapley additive explanation method (SHAP) was used to conduct early and interpretable predictions of AKI. Results The analysis included 938 eligible patients with RM. The XGBoost model exhibited superior performance (area under the receiver operating characteristic curve [AUC] = 0.767) compared to the other models (logistic regression, AUC = 0.711; support vector machine, AUC = 0.693; random forest, AUC = 0.728; and naive Bayesian, AUC = 0.700). Conclusion Although the XGBoost model performance could be improved from an absolute perspective, it provides better predictive performance than other models for estimating the AKI in patients with RM based on patient characteristics in the first 24 hours after admission to an intensive care unit. Furthermore, including SHAP to elucidate AKI-related factors enables individualized patient treatment, potentially leading to improved prognoses for patients with RM.

Assessment of EMR ML Mining Methods for Measuring Association between Metal Mixture and Mortality for Hypertension.

There are limited data available regarding the connection between heavy metal exposure and mortality among hypertension patients. We intend to establish an interpretable machine learning (ML) model with high efficiency and robustness that monitors mortality based on heavy metal exposure among hypertension patients. Our datasets were obtained from the US National Health and Nutrition Examination Survey (NHANES, 2013-2018). We developed 5 ML models for mortality prediction among hypertension patients by heavy metal exposure, and tested them by 10 discrimination characteristics. Further, we chose the optimally performing model after parameter adjustment by genetic algorithm (GA) for prediction. Finally, in order to visualize the model's ability to make decisions, we used SHapley Additive exPlanation (SHAP) and Local Interpretable Model-Agnostic Explanations (LIME) algorithm to illustrate the features. The study included 2347 participants in total. A best-performing eXtreme Gradient Boosting (XGB) with GA for mortality prediction among hypertension patients by 13 heavy metals was selected (AUC 0.959; 95% CI 0.953-0.965; accuracy 96.8%). According to sum of SHAP values, cadmium (0.094), cobalt (2.048), lead (1.12), tungsten (0.129) in urine, and lead (2.026), mercury (1.703) in blood positively influenced the model, while barium (-0.001), molybdenum (-2.066), antimony (-0.398), tin (-0.498), thallium (-2.297) in urine, and selenium (-0.842), manganese (-1.193) in blood negatively influenced the model. Hypertension patients' mortality associated with heavy metal exposure was predicted by an efficient, robust, and interpretable GA-XGB model with SHAP and LIME. Cadmium, cobalt, lead, tungsten in urine, and mercury in blood are positively correlated with mortality, while barium, molybdenum, antimony, tin, thallium in urine, and lead, selenium, manganese in blood is negatively correlated with mortality.

A Comparison of Interpretable Machine Learning Approaches to Identify Outpatient Clinical Phenotypes Predictive of First Acute Myocardial Infarction.

Acute myocardial infarctions are deadly to patients and burdensome to healthcare systems. Most recorded infarctions are patients' first, occur out of the hospital, and often are not accompanied by cardiac comorbidities. The clinical manifestations of the underlying pathophysiology leading to an infarction are not fully understood and little effort exists to use explainable machine learning to learn predictive clinical phenotypes before hospitalization is needed. We extracted outpatient electronic health record data for 2641 case and 5287 matched-control patients, all without pre-existing cardiac diagnoses, from the Michigan Medicine Health System. We compare six different interpretable, feature extraction approaches, including temporal computational phenotyping, and train seven interpretable machine learning models to predict the onset of first acute myocardial infarction within six months. Using temporal computational phenotypes significantly improved the model performance compared to alternative approaches. The mean cross-validation test set performance exhibited area under the receiver operating characteristic curve values as high as 0.674. The most consistently predictive phenotypes of a future infarction include back pain, cardiometabolic syndrome, family history of cardiovascular diseases, and high blood pressure. Computational phenotyping of longitudinal health records can improve classifier performance and identify predictive clinical concepts. State-of-the-art interpretable machine learning approaches can augment acute myocardial infarction risk assessment and prioritize potential risk factors for further investigation and validation.

Machine-Learning Prediction of Curie Temperature from Chemical Compositions of Ferromagnetic Materials.

Room-temperature ferromagnets are high-value targets for discovery given the ease by which they could be embedded within magnetic devices. However, the multitude of potential interactions among magnetic ions and their surrounding environments renders the prediction of thermally stable magnetic properties challenging. Therefore, it is vital to explore methods that can effectively screen potential candidates to expedite the discovery of novel ferromagnetic materials within highly intricate feature spaces. To this end, we explore machine-learning (ML) methods as a means to predict the Curie temperature (Tc) of ferromagnetic materials by discerning patterns within materials databases. This study emphasizes the importance of feature analysis and selection in ML modeling and demonstrates the efficacy of our gradient-boosted statistical feature-selection workflow for training predictive models. The models are fine-tuned through Bayesian optimization, using features derived solely from the chemical compositions of the materials data, before the model predictions are evaluated against literature values. We have collated ca. 35,000 Tc values and the performance of our workflow is benchmarked against state-of-the-art algorithms, the results of which demonstrate that our methodology is superior to the majority of alternative methods. In a 10-fold cross-validation, our regression model realized an R2 of (0.92 ± 0.01), an MAE of (40.8 ± 1.9) K, and an RMSE of (80.0 ± 5.0) K. We demonstrate the utility of our ML model through case studies that forecast Tc values for rare-earth intermetallic compounds and generate magnetic phase diagrams for various chemical systems. These case studies highlight the importance of a systematic approach to feature analysis and selection in enhancing both the predictive capability and interpretability of ML models, while being devoid of potential human bias. They demonstrate the advantages of such an approach over a mere reliance on algorithmic complexity and a black-box treatment in ML-based modeling within the domain of computational materials science.

A Machine Learning Model for Risk Stratification of Postdiagnosis Diabetic Ketoacidosis Hospitalization in Pediatric Type 1 Diabetes: Retrospective Study.

Diabetic ketoacidosis (DKA) is the leading cause of morbidity and mortality in pediatric type 1 diabetes (T1D), occurring in approximately 20% of patients, with an economic cost of $5.1 billion/year in the United States. Despite multiple risk factors for postdiagnosis DKA, there is still a need for explainable, clinic-ready models that accurately predict DKA hospitalization in established patients with pediatric T1D. We aimed to develop an interpretable machine learning model to predict the risk of postdiagnosis DKA hospitalization in children with T1D using routinely collected time-series of electronic health record (EHR) data. We conducted a retrospective case-control study using EHR data from 1787 patients from among 3794 patients with T1D treated at a large tertiary care US pediatric health system from January 2010 to June 2018. We trained a state-of-the-art; explainable, gradient-boosted ensemble (XGBoost) of decision trees with 44 regularly collected EHR features to predict postdiagnosis DKA. We measured the model's predictive performance using the area under the receiver operating characteristic curve-weighted F1-score, weighted precision, and recall, in a 5-fold cross-validation setting. We analyzed Shapley values to interpret the learned model and gain insight into its predictions. Our model distinguished the cohort that develops DKA postdiagnosis from the one that does not (P<.001). It predicted postdiagnosis DKA risk with an area under the receiver operating characteristic curve of 0.80 (SD 0.04), a weighted F1-score of 0.78 (SD 0.04), and a weighted precision and recall of 0.83 (SD 0.03) and 0.76 (SD 0.05) respectively, using a relatively short history of data from routine clinic follow-ups post diagnosis. On analyzing Shapley values of the model output, we identified key risk factors predicting postdiagnosis DKA both at the cohort and individual levels. We observed sharp changes in postdiagnosis DKA risk with respect to 2 key features (diabetes age and glycated hemoglobin at 12 months), yielding time intervals and glycated hemoglobin cutoffs for potential intervention. By clustering model-generated Shapley values, we automatically stratified the cohort into 3 groups with 5%, 20%, and 48% risk of postdiagnosis DKA. We have built an explainable, predictive, machine learning model with potential for integration into clinical workflow. The model risk-stratifies patients with pediatric T1D and identifies patients with the highest postdiagnosis DKA risk using limited follow-up data starting from the time of diagnosis. The model identifies key time points and risk factors to direct clinical interventions at both the individual and cohort levels. Further research with data from multiple hospital systems can help us assess how well our model generalizes to other populations. The clinical importance of our work is that the model can predict patients most at risk for postdiagnosis DKA and identify preventive interventions based on mitigation of individualized risk factors.

Advancing credit risk modelling with Machine Learning: A comprehensive review of the state-of-the-art

Ensuring financial stability necessitates responsible credit granting, so lending institutions maintain sufficient regulatory capital to withstand losses from defaults. Classification methods like Logistic Regression (LR) have long been the standard for estimating default likelihood, but they struggle with increasing operational demands and regulatory standards. Advances in Machine Learning (ML) now allow models to handle larger datasets with greater predictive power. Algorithms such as Decision Trees (DT), Support Vector Machines (SVM), and Neural Networks (NN), though not readily transferable to practical applications, have the potential to enhance the credit risk modelling process. While much of the literature on ML in credit risk focuses on sophisticated, experimental algorithms not yet ready for commercial use, few address the steps and nuances of model development. Our article fills this gap by systematically reviewing the literature on ML in consumer credit risk modelling, highlighting the current state of scientific knowledge. We analyze the frequent steps involved and demonstrate how ML's strengths can be leveraged throughout them. From pre-processing steps like data cleaning, transformation, and reduction, to variable selection and parameter optimization, our findings indicate that ML is becoming increasingly important in financial risk management. However, standardized modelling procedures are needed before complex ML classifiers can be commercially deployed. Meanwhile, although relatively interpretable ML models built around DTs and Random Forests (RF) do showcase increased performance in popular credit risk datasets, it is clear that the potential to benefit from ML spans further, with complex models enabling increased performance by aiding simpler, explainable classification algorithms.

A Visual Analytics Conceptual Framework for Explorable and Steerable Partial Dependence Analysis.

Machine learning techniques are a driving force for research in various fields, from credit card fraud detection to stock analysis. Recently, a growing interest in increasing human involvement has emerged, with the primary goal of improving the interpretability of machine learning models. Among different techniques, Partial Dependence Plots (PDP) represent one of the main model-agnostic approaches for interpreting how the features influence the prediction of a machine learning model. However, its limitations (i.e., visual interpretation, aggregation of heterogeneous effects, inaccuracy, and computability) could complicate or misdirect the analysis. Moreover, the resulting combinatorial space can be challenging to explore both computationally and cognitively when analyzing the effects of more features at the same time. This article proposes a conceptual framework that enables effective analysis workflows, mitigating state-of-the-art limitations. The proposed framework allows for exploring and refining computed partial dependences, observing incrementally accurate results, and steering the computation of new partial dependences on user-selected subspaces of the combinatorial and intractable space. With this approach, the user can save both computational and cognitive costs, in contrast with the standard monolithic approach that computes all the possible combinations of features on all their domains in batch. The framework is the result of a careful design process involving experts' knowledge during its validation and informed the development of a prototype, W4SP1, that demonstrates its applicability traversing its different paths. A case study shows the advantages of the proposed approach.

Intelligent diagnosis of Kawasaki disease from real-world data using interpretable machine learning models

ObjectiveThis study aimed to leverage real-world electronic medical record data to develop interpretable machine learning models for diagnosis of Kawasaki disease while also exploring and prioritizing the significant risk factors. MethodsA comprehensive study was conducted on 4087 pediatric patients at the Children’s Hospital of Chongqing, China. The study collected demographic data, physical examination results, and laboratory findings. Statistical analyses were performed using IBM SPSS Statistics, Version 26.0. The optimal feature subset was used to develop intelligent diagnostic prediction models based on the Light Gradient Boosting Machine, Explainable Boosting Machine (EBM), Gradient Boosting Classifier (GBC), Fast Interpretable Greedy-Tree Sums, Decision Tree, AdaBoost Classifier, and Logistic Regression. Model performance was evaluated in three dimensions: discriminative ability via receiver operating characteristic curves, calibration accuracy using calibration curves, and interpretability through SHAP (SHapley Additive exPlanations) and LIME (Local Interpretable Model-Agnostic Explanations). ResultsIn this study, Kawasaki disease was diagnosed in 2971 participants. Analysis was conducted on 31 indicators, including red blood cell distribution width and erythrocyte sedimentation rate. The EBM model demonstrated superior performance relative to other models, with an area under the curve of 0.97, second only to the GBC model. Furthermore, the EBM model exhibited the highest calibration accuracy and maintained its interpretability without relying on external analytical tools such as SHAP and LIME, thus reducing interpretation biases. Platelet distribution width, total protein, and erythrocyte sedimentation rate were identified by the model as significant predictors for the diagnosis of Kawasaki disease. ConclusionThis study used diverse machine learning models for early diagnosis of Kawasaki disease. The findings demonstrated that interpretable models such as EBM outperformed traditional machine learning models in terms of both interpretability and performance. Ensuring consistency between predictive models and clinical evidence is crucial for the successful integration of artificial intelligence into real-world clinical practice.

Enhancing cardiovascular risk assessment with advanced data balancing and domain knowledge-driven explainability

In medical risk prediction, such as predicting heart disease, machine learning (ML) classifiers must achieve high accuracy, precision, and recall to minimize the chances of incorrect diagnoses or treatment recommendations. However, real-world datasets often have imbalanced data, which can affect classifier performance. Traditional data balancing methods can lead to overfitting and underfitting, making it difficult to identify potential health risks accurately. Early prediction of heart attacks is of paramount importance, and researchers have developed ML-based systems to address this problem. However, much of the existing ML research is based on a single dataset, often ignoring performance evaluation across multiple datasets. As the demand for interpretable ML models grows, model interpretability becomes central to revealing insights and feature effects within predictive models. To address these challenges, we present a novel data balancing technique that uses a divide-and-conquer strategy with the K-Means clustering algorithm to segment the dataset. The performance of our approach is highlighted through comparisons with established techniques, which demonstrate the superiority of our proposed method. To address the challenge of inter-dataset discrepancies, we use two different datasets. Our holistic pipeline, strengthened by the innovative balancing technique, effectively addresses performance discrepancies, culminating in a significant improvement from 81% to 90%. Furthermore, through advanced statistical analysis, it has been determined that the 95% confidence interval for the AUC metric of our method ranges from 0.8187 to 0.8411. This observation serves to underscore the consistency and reliability of our approach, demonstrating its ability to achieve high performance across a range of scenarios. Incorporating Explainable AI (XAI), we examine the feature rankings and their contributions within the best performing Random Forest model. While the domain expert feedback is consistent with the explanatory power of XAI, some differences remain. Nevertheless, a remarkable convergence in feature ranking and weighting is observed, bridging the insights from XAI tools and domain expert perspectives.

Industry return prediction via interpretable deep learning

We apply an interpretable machine learning model, the LassoNet, to forecast and trade U.S. industry portfolio returns. The model combines a regularization mechanism with a neural network architecture. A cooperative game-theoretic algorithm is also applied to interpret our findings. The latter hierarchizes the covariates based on their contribution to the overall model performance. Our findings reveal that the LassoNet outperforms various linear and nonlinear benchmarks concerning out-of-sample forecasting accuracy and provides economically meaningful and profitable predictions. Valuation ratios are the most crucial covariates, followed by individual and cross-industry lagged returns. The constructed industry ETF portfolios attain positive Sharpe ratios and positive and statistically significant alphas, surviving even transaction costs.

Strong Optimal Classification Trees

Decision trees are among the most interpretable and popular machine learning models that are used routinely in applications ranging from revenue management to medicine. Traditional heuristic methods, although fast, lack modeling flexibility for incorporating constraints such as fairness and do not guarantee optimality. Recent efforts aim to overcome these limitations using mixed-integer optimization (MIO) for better modeling flexibility and optimality, but speed remains an issue. In “Strong Optimal Classification Trees,” Aghaei, Gómez, and Vayanos use integer optimization and polyhedral theory to create an MIO-based formulation with a strong LO relaxation resulting in a 29% speed-up in training time compared with state-of-the-art MIO-based formulations, as well as up to an 8% improvement in out-of-sample accuracy.

Harnessing machine learning for predictive maintenance in energy infrastructure: A review of challenges and solutions

Predictive maintenance (PdM) has emerged as a vital strategy for optimizing the reliability and efficiency of energy infrastructure. In this paper, we present a comprehensive review of the challenges and solutions associated with harnessing machine learning (ML) techniques for predictive maintenance in the energy sector. The adoption of ML algorithms in predictive maintenance holds immense promise for mitigating equipment failures, reducing downtime, and optimizing maintenance schedules. However, several challenges impede the effective implementation of ML-based PdM strategies. These challenges include the need for large and high-quality data sets, the complexity of integrating heterogeneous data sources, and the interpretability of ML models in real-world settings. To address these challenges, we discuss various solutions and best practices. These include data preprocessing techniques to handle noisy and incomplete data, feature engineering methods for extracting meaningful insights, and model interpretability approaches for enhancing trust and understanding of ML predictions. Additionally, we explore the integration of domain knowledge and human expertise into ML algorithms to improve predictive accuracy and relevance. Furthermore, we examine the role of edge computing and distributed ML techniques in enabling real-time predictive maintenance, particularly in remote or resource-constrained environments. We also discuss the importance of regulatory compliance, privacy protection, and ethical considerations in the deployment of ML-based PdM solutions.