Development Of Machine Learning Models Research Articles

Towards inclusivity in AI: A comparative study of cognitive engagement between marginalized female students and peers

AbstractThis study addresses the need for inclusive AI education by focusing on marginalized female students who historically lack access to learning opportunities in computing. It applies the theoretical framework of intersectionality to understand how gender, race and ethnicity intersect to shape these students' learning experiences and outcomes. Specifically, this study investigated 27 high‐school students' cognitive engagement in machine learning practices. We conducted the Wilcoxon–Mann–Whitney test to explore differences in cognitive engagement between marginalized female students and their peers, employed comparative content analysis to delve into significant differences and analysed interview data thematically to gain deeper insights into students' machine learning model development processes. The findings indicated that, when engaging in machine learning practices requiring drawing diverse cultural perspectives, marginalized female students demonstrated significantly higher performance compared to their peers. In particular, marginalized female students exhibited strengths in holistic language analysis, paying attention to writers' intentions and recognizing cultural nuances in language. This study suggests that integrating language analysis and machine learning across subjects has the potential to empower marginalized female students and amplify their perspectives. Furthermore, it calls for a strengths‐based approach to reshape the narrative of underrepresentation and promote equitable participation in machine learning and AI. Practitioner notesWhat is already known about this topic Female students, particularly those from underrepresented groups such as African American and Latina students, often experience low levels of cognitive engagement in computing. Marginalized female students possess unique strengths that, when nurtured, have the potential to not only transform their own learning experiences but also contribute to the advancement of the computing field. It is critical to empower marginalized female students in K‐12 AI (ie, a subfield of computing) education, seeking to bridge the gender and racial disparity in AI. What this paper adds Marginalized female students outperformed their peers in responding to machine learning questions related to feature analysis and feature distribution interpretation. When responding to these questions, they demonstrated a holistic approach to analysing language by considering interactions between features and writers' intentions. They drew on knowledge about how language was used to convey meaning in different cultural contexts. Implications for practice and/or policy Educators should design learning environments that encourage students to draw upon their cultural backgrounds, linguistic insights and diverse experiences to enhance their engagement and performance in AI‐related activities. Educators should strategically integrate language analysis and machine learning across different subjects to create interdisciplinary learning experiences that support students' exploration of the interplay among language, culture and AI. Educational institutions and policy initiatives should adopt a strengths‐based approach that focuses on empowering marginalized female students by acknowledging their inherent abilities and diverse backgrounds.

Machine-learning-based models to predict cardiovascular risk using oculomics and clinic variables in KNHANES

BackgroundRecent researches have found a strong correlation between the triglyceride-glucose (TyG) index or the atherogenic index of plasma (AIP) and cardiovascular disease (CVD) risk. However, there is a lack of research on non-invasive and rapid prediction of cardiovascular risk. We aimed to develop and validate a machine-learning model for predicting cardiovascular risk based on variables encompassing clinical questionnaires and oculomics.MethodsWe collected data from the Korean National Health and Nutrition Examination Survey (KNHANES). The training dataset (80% from the year 2008 to 2011 KNHANES) was used for machine learning model development, with internal validation using the remaining 20%. An external validation dataset from the year 2012 assessed the model’s predictive capacity for TyG-index or AIP in new cases. We included 32122 participants in the final dataset. Machine learning models used 25 algorithms were trained on oculomics measurements and clinical questionnaires to predict the range of TyG-index and AIP. The area under the receiver operating characteristic curve (AUC), accuracy, precision, recall, and F1 score were used to evaluate the performance of our machine learning models.ResultsBased on large-scale cohort studies, we determined TyG-index cut-off points at 8.0, 8.75 (upper one-third values), 8.93 (upper one-fourth values), and AIP cut-offs at 0.318, 0.34. Values surpassing these thresholds indicated elevated cardiovascular risk. The best-performing algorithm revealed TyG-index cut-offs at 8.0, 8.75, and 8.93 with internal validation AUCs of 0.812, 0.873, and 0.911, respectively. External validation AUCs were 0.809, 0.863, and 0.901. For AIP at 0.34, internal and external validation achieved similar AUCs of 0.849 and 0.842. Slightly lower performance was seen for the 0.318 cut-off, with AUCs of 0.844 and 0.836. Significant gender-based variations were noted for TyG-index at 8 (male AUC=0.832, female AUC=0.790) and 8.75 (male AUC=0.874, female AUC=0.862) and AIP at 0.318 (male AUC=0.853, female AUC=0.825) and 0.34 (male AUC=0.858, female AUC=0.831). Gender similarity in AUC (male AUC=0.907 versus female AUC=0.906) was observed only when the TyG-index cut-off point equals 8.93.ConclusionWe have established a simple and effective non-invasive machine learning model that has good clinical value for predicting cardiovascular risk in the general population.

Predictive model for a product without history using LightGBM. Pricing model for a new product

The article focuses on developing a predictive product pricing model using LightGBM. Also, the goal was to adapt the LightGBM method for regression problems and, especially, in the problems of forecasting the price of a product without history, that is, with a cold start.The article contains the necessary concepts to understand the working principles of the light gradient boosting machine, such as decision trees, boosting, random forests, gradient descent, GBM (Gradient Boosting Machine), GBDT (Gradient Boosting Decision Trees). The article provides detailed insights into the algorithms used for identifying split points, with a focus on the histogram-based approach.LightGBM enhances the gradient boosting algorithm by introducing an automated feature selection mechanism and giving special attention to boosting instances characterized by more substantial gradients. This can lead to significantly faster training and improved prediction performance. The Gradient-based One-Side Sampling (GOSS) and Exclusive Feature Bundling (EFB) techniques used as enhancements to LightGBM are vividly described. The article presents the algorithms for both techniques and the complete LightGBM algorithm.This work contains an experimental result. To test the lightGBM, a real dataset of one Japanese C2C marketplace from the Kaggle site was taken. In the practical part, a performance comparison between LightGBM and XGBoost (Extreme Gradient Boosting Machine) was performed. As a result, only a slight increase in estimation performance (RMSE, MAE, R-squard) was found by applying LightGBM over XGBoost, however, there exists a notable contrast in the training procedure’s time efficiency. LightGBM exhibits an almost threefold increase in speed compared to XGBoost, making it a superior choice for handling extensive datasets.This article is dedicated to the development and implementation of machine learning models for product pricing using LightGBM. The incorporation of automatic feature selection, a focus on highgradient examples, and techniques like GOSS and EFB demonstrate the model’s versatility and efficiency. Such predictive models will help companies improve their pricing models for a new product. The speed of obtaining a forecast for each element of the database is extremely relevant at a time of rapid data accumulation.

Performance optimization of Bloch surface wave based devices using an XGBoost machine learning model

Bloch surface wave (BSW) based photonic devices have attracted significant interest for bio-sensing, spectroscopy, and light manipulation applications due to their ability to confine light at dielectric interfaces. However, optimizing the geometry of BSW structures to achieve desired optical properties can be computationally expensive using conventional simulation methods for multi-parameter design spaces. In this work, we develop machine learning models based on the gradient boosting algorithm XGBoost to predict key optical characteristics of BSW devices and expedite the design process. Finite element method simulations are used to generate a dataset relating BSW structures’ excitation angle, sensitivity, and spectral response to their geometric parameters, including thickness, porosity, and surrounding refractive index. This dataset trains and validates different XGBoost regression models for photonic structure optimization. Our results demonstrate that a model utilizing deep decision trees achieves the highest predictive accuracy, with a mean absolute error of 0.09° in estimating the excitation angle for new structures. We apply this optimized model to uncover the thickness-porosity combinations, enabling a maximum sensitivity of 171-degree/RIU. This machine learning approach provides a powerful tool for the inverse design and performance enhancement of BSW photonic structures beyond the capabilities of conventional simulation-based optimization.

Biomolecular Adsorption on Nanomaterials: Combining Molecular Simulations with Machine Learning.

Adsorption free energies of 32 small biomolecules (amino acids side chains, fragments of lipids, and sugar molecules) on 33 different nanomaterials, computed by the molecular dynamics - metadynamics methodology, have been analyzed using statistical machine learning approaches. Multiple unsupervised learning algorithms (principal component analysis, agglomerative clustering, and K-means) as well as supervised linear and nonlinear regression algorithms (linear regression, AdaBoost ensemble learning, artificial neural network) have been applied. As a result, a small set of biomolecules has been identified, knowledge of adsorption free energies of which to a specific nanomaterial can be used to predict, within the developed machine learning model, adsorption free energies of other biomolecules. Furthermore, the methodology of grouping of nanomaterials according to their interactions with biomolecules has been presented.

An interpretable framework for modeling global solar radiation using tree-based ensemble machine learning and Shapley additive explanations methods

Machine learning techniques provide an effective and cost-efficient solution for estimating solar radiation for solar energy utilization. However, the reported machine learning-based solar radiation models fail to offer comprehensive explanations for their outputs. Therefore, this study aims to tackle this issue by developing machine learning models that are both accurate and interpretable. To achieve the objective, this study evaluated the performance of tree-based ensemble algorithms, using optimized combinations of model input parameters for different climate zones in China. The results showed that the extreme gradient boosting (XGBoost) models demonstrated the highest overall accuracy, model stability, and generalization ability. At the national scale, the developed XGBoost models yielded an average R2, MAE, and RMSE of 0.939, 1.226 MJ/m2, and 1.663 MJ/m2, respectively, showing significant improvements of 2.13–27.78% in RMSE compared to recently reported models. Most importantly, the state-of-the-art SHapley Additive exPlanations (SHAP) technique was integrated with the developed XGBoost models to enhance model interpretability in terms of global and local feature importance, as well as the interaction effects between model features. The results of the SHAP value analysis demonstrated the robustness of sunshine duration in modeling global solar radiation, revealing thresholds where its values undergo a shift from negative to positive effects on model output. SHAP interaction values illustrated the interaction effects among features in the developed solar radiation model, uncovering the model's complex non-linear relationships. Additionally, this study provided explanations for individual instances based on the SHAP method. Overall, this study provided an accurate, reliable, and transparent machine learning model and an enlightening framework for modeling global solar radiation at sites without observations.

Development and performance assessment of novel machine learning models for predicting postoperative pneumonia in aneurysmal subarachnoid hemorrhage patients: external validation in MIMIC-IV.

Postoperative pneumonia (POP) is one of the primary complications after aneurysmal subarachnoid hemorrhage (aSAH) and is associated with postoperative mortality, extended hospital stay, and increased medical fee. Early identification of pneumonia and more aggressive treatment can improve patient outcomes. We aimed to develop a model to predict POP in aSAH patients using machine learning (ML) methods. This internal cohort study included 706 patients with aSAH undergoing intracranial aneurysm embolization or aneurysm clipping. The cohort was randomly split into a train set (80%) and a testing set (20%). Perioperative information was collected from participants to establish 6 machine learning models for predicting POP after surgical treatment. The area under the receiver operating characteristic curve (AUC), precision-recall curve were used to assess the accuracy, discriminative power, and clinical validity of the predictions. The final model was validated using an external validation set of 97 samples from the Medical Information Mart for Intensive Care IV (MIMIC-IV) database. In this study, 15.01% of patients in the training set and 12.06% in the testing set with POP after underwent surgery. Multivariate logistic regression analysis showed that mechanical ventilation time (MVT), Glasgow Coma Scale (GCS), Smoking history, albumin level, neutrophil-to-albumin Ratio (NAR), c-reactive protein (CRP)-to-albumin ratio (CAR) were independent predictors of POP. The logistic regression (LR) model presented significantly better predictive performance (AUC: 0.91) than other models and also performed well in the external validation set (AUC: 0.89). A machine learning model for predicting POP in aSAH patients was successfully developed using a machine learning algorithm based on six perioperative variables, which could guide high-risk POP patients to take appropriate preventive measures.

Integrating mid-infrared spectroscopy, machine learning, and graphical bias correction for fatty acid prediction in water buffalo milk.

Buffalo milk, constituting 15% of global production, has higher fatty acids content than Holstein milk. Fourier-transform mid-infrared (FT-MIR) spectroscopy is widely used for dairy analysis, but its application to buffalo milk, with larger fat globules, remains understudied. The ultimate goal of this study is to develop machine learning models based on FT-MIR spectroscopy for predicting fatty acids in buffalo milk and to assess the accuracy of commercial milk analyzers. This research provides a convenient, fast, and environmentally friendly method for detecting the fatty acid composition in buffalo milk. We employed six machine learning algorithms to establish a detection model for 34 fatty acids in buffalo milk. The predictive models demonstrated robust capabilities for high-content fatty acids [C14:0, C15:0, C16:0, C17:0, C18:0, C18:1, saturated fatty acid (SFA), monounsaturated fatty acid (MUFA)], with errors within a 15% range. Traditional FT6000 detection methods exhibited limitations in measuring SFAs and polyunsaturated fatty acids (PUFA). Implementing a mean difference correction of 0.21 for MUFAs and applying regression equations (SFA × 1.0639 + 0.0705; PUFA × 0.5472 + 0.0047) significantly improved measurement accuracy. This study successfully developed a predictive model for fatty acids in Mediterranean buffalo milk based on FT-MIR spectroscopy. Additionally, a correction was applied to the existing measurement device, FT6000, enabling more accurate measurements of fatty acids in buffalo milk. The findings have practical implications for the food industry, offering a faster and more reliable approach to assess and monitor fatty acid composition in buffalo milk, potentially influencing product development and quality control processes. © 2024 Society of Chemical Industry.

Development and validation of machine learning-based models for predicting healthcare-associated bacterial/fungal infections among COVID-19 inpatients: a retrospective cohort study

BackgroundCOVID-19 and bacterial/fungal coinfections have posed significant challenges to human health. However, there is a lack of good tools for predicting coinfection risk to aid clinical work.ObjectiveWe aimed to investigate the risk factors for bacterial/fungal coinfection among COVID-19 patients and to develop machine learning models to estimate the risk of coinfection.MethodsIn this retrospective cohort study, we enrolled adult inpatients confirmed with COVID-19 in a tertiary hospital between January 1 and July 31, 2023, in China and collected baseline information at admission. All the data were randomly divided into a training set and a testing set at a ratio of 7:3. We developed the generalized linear and random forest models for coinfections in the training set and assessed the performance of the models in the testing set. Decision curve analysis was performed to evaluate the clinical applicability.ResultsA total of 1244 patients were included in the training cohort with 62 healthcare-associated bacterial/fungal infections, while 534 were included in the testing cohort with 22 infections. We found that patients with comorbidities (diabetes, neurological disease) were at greater risk for coinfections than were those without comorbidities (OR = 2.78, 95%CI = 1.61–4.86; OR = 1.93, 95%CI = 1.11–3.35). An indwelling central venous catheter or urinary catheter was also associated with an increased risk (OR = 2.53, 95%CI = 1.39–4.64; OR = 2.28, 95%CI = 1.24–4.27) of coinfections. Patients with PCT > 0.5 ng/ml were 2.03 times (95%CI = 1.41–3.82) more likely to be infected. Interestingly, the risk of coinfection was also greater in patients with an IL-6 concentration < 10 pg/ml (OR = 1.69, 95%CI = 0.97–2.94). Patients with low baseline creatinine levels had a decreased risk of bacterial/fungal coinfections(OR = 0.40, 95%CI = 0.22–0.71). The generalized linear and random forest models demonstrated favorable receiver operating characteristic curves (ROC = 0.87, 95%CI = 0.80–0.94; ROC = 0.88, 95%CI = 0.82–0.93) with high accuracy, sensitivity and specificity of 0.86vs0.75, 0.82vs0.86, 0.87vs0.74, respectively. The corresponding calibration evaluation P statistics were 0.883 and 0.769.ConclusionsOur machine learning models achieved strong predictive ability and may be effective clinical decision-support tools for identifying COVID-19 patients at risk for bacterial/fungal coinfection and guiding antibiotic administration. The levels of cytokines, such as IL-6, may affect the status of bacterial/fungal coinfection.

An Interpretable Machine Learning Approach to Predict Survival Outcomes in Spinal and Sacropelvic Chordomas.

Retrospective, population-based cohort study. This study aimed to develop machine learning (ML) models to predict five-year and 10-year mortality in spinal and sacropelvic chordoma patients and integrate them into a web application for enhanced prognostication. Past research has uncovered factors influencing survival in spinal chordoma patients. While identifying individual predictors is important, personalized survival predictions are equally vital. Though prior efforts have resulted in nomograms aiming to serve this purpose, they cannot capture complex interactions within data and rely on statistical assumptions that may not fit real-world data. Adult spinal and sacropelvic chordoma patients were identified from the National Cancer Database. Sociodemographic, clinicopathologic, diagnostic, and treatment-related variables were utilized as predictive features. Five supervised ML algorithms (TabPFN, CatBoost, XGBoost, LightGBM, and Random Forest) were implemented to predict mortality at five and 10 years postdiagnosis. Model performance was primarily evaluated using the area under the receiver operating characteristic (AUROC). SHapley Additive exPlanations (SHAP) values and partial dependence plots provided feature importance and interpretability. The top models were integrated into a web application. From the NCDB, 1206 adult patients diagnosed with histologically confirmed spinal and sacropelvic chordomas were retrieved for the five-year mortality outcome [423 (35.1%) with five-year mortality] and 801 patients for the 10-year mortality outcome [588 (73.4%) with 10-year mortality]. Top-performing models for both of the outcomes were the models created with the CatBoost algorithm. The CatBoost model for five-year mortality predictions displayed a mean AUROC of 0.801, and the CatBoost model predicting 10-year mortality yielded a mean AUROC of 0.814. This study developed ML models that can accurately predict five-year to 10-year survival probabilities in spinal chordoma patients. Integrating these interpretable, personalized prognostic models into a web application provides quantitative survival estimates for a given patient. The local interpretability enables transparency into how predictions are influenced. Further external validation is warranted to support generalizability and clinical utility.

Advanced stacked integration method for forecasting long-term drought severity: CNN with machine learning models

Study regionEight governorates in upper Egypt namely Aswan, Asyut, Beni-Suef, Fayoum, Luxor, Minya, Qena and Sohag. Study focusThis study aims to develop novel hybrid machine learning (ML) models for forecasting the drought phenomena based on limited inputs for the eight Egyptian govern-orates, and ii) evaluate the performance and accuracy of the developed ML models for predicting Palmer Drought Severity Index (PDSI) to recommend the optimal model based on performance statistical metrics. The hybrid ML models were Convolution Neural Networks (CNN)-Long Short-Term Memory (LSTM), CNN-Random Forest (RF), CNN-Support Vector Machine (SVR), and CNN-Extreme Gradient Boosting (XGB). New hydrological insights for the regionResults showed that CNN-LSTM model outperformed the others followed by CNN-RF. Values of NSE, MAE, MARE, IA, R2, and RMSE for CNN-LSTM were 0.885, 0.915, − 2.073, 0.967, 0.885, and 0.573, respectively. For the testing stage CNN-SVR model was found to perform the best; average values of NSE, MAE, MARE, IA, R2, and RMSE were 0.828, 0.364, − 2.903, 0.950, 0.828 and 0.688, respectively. This study provided a way forward for convenient estimation of the PDSI Index from the meteorological data in terms of advancing deep learning algorithms. The developed hybrid models, more or less, can satisfactory predict PDSI values. Additionally, the study suggests the CNN-LSTM model as the most suitable model to advance future investigation in the study area.

Development and validation of machine learning models to predict frailty risk for elderly.

Early identification and intervention of the frailty of the elderly will help lighten the burden of social medical care and improve the quality of life of the elderly. Therefore, we used machine learning (ML) algorithm to develop models to predict frailty risk in the elderly. A prospective cohort study. We collected data on 6997 elderly people from Chinese Longitudinal Healthy Longevity Study wave 6-7 surveys (2011-2012, 2014). After the baseline survey in 1998 (wave 1), the project conducted follow-up surveys (wave 2-8) in 2000-2018. The osteoporotic fractures index was used to assess frailty. Four ML algorithms (random forest [RF], support vector machine, XGBoost and logistic regression [LR]) were used to develop models to identify the risk factors of frailty and predict the risk of frailty. Different ML models were used for the prediction of frailty risk in the elderly and frailty risk was trained on a cohort of 4385 elderly people with frailty (split into a training cohort [75%] and internal validation cohort [25%]). The best-performing model for each study outcome was tested in an external validation cohort of 6997 elderly people with frailty pooled from the surveys (wave 6-7). Model performance was assessed by receiver operating curve and F2-score. Among the four ML models, the F2-score values were similar (0.91 vs. 0.91 vs. 0.88 vs. 0.90), and the area under the curve (AUC) values of RF model was the highest (0.75), followed by LR model (0.74). In the final two models, the AUC values of RF and LR model were similar (0.77 vs. 0.76) and their accuracy was identical (87.4% vs. 87.4%). Our study developed a preliminary prediction model based on two different ML approaches to help predict frailty risk in the elderly. The presented models from this study can be used to inform healthcare providers to predict the frailty probability among older adults and maybe help guide the development of effective frailty risk management interventions. Detecting frailty at an early stage and implementing timely targeted interventions may help to improve the allocation of health care resources and to reduce frailty-related burden. Identifying risk factors for frailty could be beneficial to provide tailored and personalized care intervention for older adults to more accurately prevent or improve their frail conditions so as to improve their quality of life. The study has adhered to STROBE guidelines. No patient or public contribution.

Synthetic Force-Field Database for Training Machine Learning Models to Predict Mobility-Preserving Coarse-Grained Molecular-Simulation Potentials.

Balancing accuracy and efficiency is a common problem in molecular simulation. This tradeoff is evident in coarse-grained molecular dynamics simulation, which prioritizes efficiency, and all-atom molecular simulation, which prioritizes accuracy. Despite continuous efforts, creating a coarse-grained model that accurately captures both the system's structure and dynamics remains elusive. In this article, we present a data-driven approach for constructing coarse-grained models that aim to describe both the structure and dynamics of the system equally well. While the development of machine learning models is well-received in the scientific community, the significance of dataset creation for these models is often overlooked. However, data-driven approaches cannot progress without a robust dataset. To address this, we construct a database of synthetic coarse-grained potentials generated from unphysical all-atom models. A neural network is trained with the generated database to predict the coarse-grained potentials of real liquids. We evaluate their quality by calculating the combined loss of structural and dynamical accuracy upon coarse-graining. When we compare our machine learning-based coarse-grained potential with the one from iterative Boltzmann inversion, the machine learning prediction turns out better for all eight hydrocarbon liquids we studied. As all-atom surfaces turn more nonspherical, both ways of coarse-graining degrade. Still, the neural network outperforms iterative Boltzmann inversion in constructing good quality coarse-grained models for such cases. The synthetic database and the developed machine learning models are freely available to the community, and we believe that our approach will generate interest in efficiently deriving accurate coarse-grained models for liquids.

Predicting the complexity and mortality of polytrauma patients with machine learning models

We aim to develop machine learning (ML) models for predicting the complexity and mortality of polytrauma patients using clinical features, including physician diagnoses and physiological data. We conducted a retrospective analysis of a cohort comprising 756 polytrauma patients admitted to the intensive care unit (ICU) at Pizhou People’s Hospital Trauma Center, Jiangsu, China between 2020 and 2022. Clinical parameters encompassed demographics, vital signs, laboratory values, clinical scores and physician diagnoses. The two primary outcomes considered were mortality and complexity. We developed ML models to predict polytrauma mortality or complexity using four ML algorithms, including Support Vector Machine (SVM), Random Forest (RF), Artificial Neural Network (ANN) and eXtreme Gradient Boosting (XGBoost). We assessed the models’ performance and compared the optimal ML model against three existing trauma evaluation scores, including Injury Severity Score (ISS), Trauma Index (TI) and Glasgow Coma Scale (GCS). In addition, we identified several important clinical predictors that made contributions to the prognostic models. The XGBoost-based polytrauma mortality prediction model demonstrated a predictive ability with an accuracy of 90% and an F-score of 88%, outperforming SVM, RF and ANN models. In comparison to conventional scoring systems, the XGBoost model had substantial improvements in predicting the mortality of polytrauma patients. External validation yielded strong stability and generalization with an accuracy of up to 91% and an AUC of 82%. To predict polytrauma complexity, the XGBoost model maintained its performance over other models and scoring systems with good calibration and discrimination abilities. Feature importance analysis highlighted several clinical predictors of polytrauma complexity and mortality, such as Intracranial hematoma (ICH). Leveraging ML algorithms in polytrauma care can enhance the prognostic estimation of polytrauma patients. This approach may have potential value in the management of polytrauma patients.

Interpretable AI models for judicial decisionmaking: beyond explicability towards legal due process

As AI algorithms are employed to apply legal rules in determining rights and obligations, questions related to the observance of due legal process arise. The development of opaque machine learning models, whose predictions cannot be satisfactorily explained, has spurred debates around the idea of explainability of AI models for decision-making. The article argues that (i) in relation to AI models for judicial decision-making, the standard of explainability, besides proving insufficient to meet the requirements for publicity and reasoning of judicial decisions, imposes a form of nakedness not required of human judges; and (ii) a more appropriate standard would be that of interpretable models for judicial decision-making, characterized as able to offer decisions that are referred to current law (legality), internally and externally coherent (consistency), and compatible with the decision of a human judge in a similar case.

Rapid and Accurate Screening of the COF Space for Natural Gas Purification: COFInformatics.

In this work, we introduced COFInformatics, a computational approach merging molecular simulations and machine learning (ML) algorithms, to evaluate all synthesized and hypothetical covalent organic frameworks (COFs) for the CO2/CH4 mixture separation under four different adsorption-based processes: pressure swing adsorption (PSA), vacuum swing adsorption (VSA), temperature swing adsorption (TSA), and pressure-temperature swing adsorption (PTSA). We first extracted structural, chemical, energy-based, and graph-based molecular fingerprint features of every single COF structure in the very large COF space, consisting of nearly 70,000 materials, and then performed grand canonical Monte Carlo simulations to calculate the CO2/CH4 mixture adsorption properties of 7540 COFs. These features and simulation results were used to develop ML models that accurately and rapidly predict CO2/CH4 mixture adsorption and separation properties of all 68,614 COFs. The most efficient separation process and the best adsorbent candidates among the entire COF spectrum were identified and analyzed in detail to reveal the most important molecular features that lead to high-performance adsorbents. Our results showed that (i) many hypoCOFs outperform synthesized COFs by achieving higher CO2/CH4 selectivities; (ii) the top COF adsorbents consist of narrow pores and linkers comprising aromatic, triazine, and halogen groups; and (iii) PTSA is the most efficient process to use COF adsorbents for natural gas purification. We believe that COFInformatics promises to expedite the evaluation of COF adsorbents for CO2/CH4 separation, thereby circumventing the extensive, time- and resource-intensive molecular simulations.

Development of a novel continuum damage mechanics‐based machine learning approach for vibration fatigue assessment of fastener clip subjected to high‐frequency vibration

AbstractThis paper proposes a novel method based on continuum damage mechanics (CDM) and machine learning (ML) models to evaluate the vibration fatigue behavior of W1‐type railway fastener clips subjected to high‐frequency vibration. Firstly, static and fatigue tests are conducted on 60Si2Mn spring steel to acquire elastic modulus, tensile strength, and P‐S‐N curves. Subsequently, a CDM model is established, and numerical simulations are performed under various working conditions to obtain the fatigue characteristics of the clips. Finally, the ML model is trained using numerical simulation results, thereby establishing a mapping model between the working conditions and fatigue characteristics. The developed ML model demonstrates high accuracy in predicting the vibration fatigue life of the clips. Moreover, the Shapley Additive Explanations (SHAP) algorithm is employed to elucidate the ML model, revealing that the vibration frequency has a greater impact on the fatigue life of the clips compared to the vibration displacement.

Enhancing Reliability of Advanced Driver-Assistance Systems through Predictive Maintenance and Data-Driven Insights

The advancement of Advanced Driver Assistance Systems (ADAS) marks a pivotal evolution in automotive technology, aiming to enhance road safety and driving efficiency through a wide array of functionalities like blind spot detection, emergency braking, and adaptive cruise control. This research paper delves into the operational integrity, performance metrics, and maintenance strategies of ADAS components, underpinned by a comprehensive methodology involving data collection, pre-processing, feature engineering, machine learning model development, and rigorous validation processes. Systematic inspection of ADAS components indicates their importance in vehicle safety and reliability. The visibility, distance, speed, and steering angle of front cameras, LiDAR, radar, and ultrasonic sensors are carefully evaluated. Maintenance logs show proactive error code management, boosting efficiency. SVM, Gradient Boosting, and Random Forest machine learning models predicted ADAS component failures during validation and testing. Random Forest scored 90% accuracy, 92% precision, 88% recall, and 90% F1. Gradient Boosting was the most accurate, with 93% accuracy, 94% precision, 91% recall, and 92% F1. SVM predicted ADAS component failures with 88% accuracy, 90% precision, 85% recall, and 87% F1 score. Machine learning helps shift from reactive to proactive maintenance. Modelling sensor signal quality, actuator reaction times, error code frequencies, and maintenance intervals enables predictive maintenance and failure detection. Feature engineering builds predictive models using maintenance logs and operational KPIs. The models predict ADAS component failures, boosting vehicle safety and dependability. Using external data improves predictive maintenance models. The maintenance model's adaptability and forecast accuracy are proved by ADAS operation after traffic, accident, and manufacturer upgrades. Predictive maintenance and machine learning improve ADAS dependability and safety, the study found. Advanced analytics and data-driven insights can reduce automotive system failures, improving safety and reliability.

Applying Recurrent Neural Networks and Blocked Cross-Validation to Model Conventional Drinking Water Treatment Processes

The jar test is the current standard method for predicting the performance of a conventional drinking water treatment (DWT) process and optimizing the coagulant dose. This test is time-consuming and requires human intervention, meaning it is infeasible for making continuous process predictions. As a potential alternative, we developed a machine learning (ML) model from historical DWT plant data that can operate continuously using real-time sensor data without human intervention for predicting clarified water turbidity 15 min in advance. We evaluated three types of models: multilayer perceptron (MLP), the long short-term memory (LSTM) recurrent neural network (RNN), and the gated recurrent unit (GRU) RNN. We also employed two training methodologies: the commonly used holdout method and the theoretically correct blocked cross-validation (BCV) method. We found that the RNN with GRU was the best model type overall and achieved a mean absolute error on an independent production set of as low as 0.044 NTU. We further found that models trained using BCV typically achieve errors equal to or lower than their counterparts trained using holdout. These results suggest that RNNs trained using BCV are superior for the development of ML models for DWT processes compared to those reported in earlier literature.

Unlocking Machine Learning Model Decisions: A Comparative Analysis of LIME and SHAP for Enhanced Interpretability

XAI is critical for establishing trust and enabling the appropriate development of machine learning models. By offering transparency into how these models make judgements, XAI enables researchers and users to uncover potential biases, admit limits, and eventually enhance the fairness and dependability of AI systems. In this paper, we demonstrates two techniques, LIME and SHAP, used to improve the interpretability of machine learning models. Assessing Explainable AI (XAI) approaches is critical in searching for transparent and interpretable artificial intelligence (AI) models. Explainable AI (XAI) approaches are designed to provide insight into how complex models make decisions. This paper thoroughly analyzes two prominent XAI methods: Shapley Additive explanations (SHAP) and Local Interpretable Model-agnostic Explanations (LIME). This study aims to understand the decision made by a machine learning model and how the model came to that decision. We discuss the approaches and framework of both LIME and SHAP and assess their behavior in predicting the model's outcome.