Pub Date : 2025-12-01Epub Date: 2025-09-27DOI: 10.1016/j.mlwa.2025.100738
Sow Thierno Hamidou, Adda Mehdi
Intrusion Detection Systems (IDS) face major challenges in network security, notably the need to combine a high detection rate with reliable performance. This reliability is often affected by class imbalances and inadequate hyperparameter optimization. This article addresses the issue of improving the detection rate of IDS by evaluating and comparing three machine learning algorithms: Random Forest (RF), XGBoost, and Deep Neural Networks (DNN), using the NSL-KDD dataset. In our methodology, we integrate SMOTE (Synthetic Minority Oversampling Technique) to tackle the unbalanced nature of the data, ensuring a more balanced representation of the different classes. This approach helps optimize model performance, reduce bias, and enhance robustness. Additionally, hyperparameter optimization is performed using Optuna, ensuring that each algorithm operates at its optimal level. The results show that our model, using the Random Forest algorithm, achieves an accuracy of 99.80%, surpassing the performance of XGBoost and Deep Neural Networks (DNN). This makes our approach a true asset for intrusion detection methods in computer networks.
{"title":"Enhancing IDS performance through a comparative analysis of Random Forest, XGBoost, and Deep Neural Networks","authors":"Sow Thierno Hamidou, Adda Mehdi","doi":"10.1016/j.mlwa.2025.100738","DOIUrl":"10.1016/j.mlwa.2025.100738","url":null,"abstract":"<div><div>Intrusion Detection Systems (IDS) face major challenges in network security, notably the need to combine a high detection rate with reliable performance. This reliability is often affected by class imbalances and inadequate hyperparameter optimization. This article addresses the issue of improving the detection rate of IDS by evaluating and comparing three machine learning algorithms: Random Forest (RF), XGBoost, and Deep Neural Networks (DNN), using the NSL-KDD dataset. In our methodology, we integrate SMOTE (Synthetic Minority Oversampling Technique) to tackle the unbalanced nature of the data, ensuring a more balanced representation of the different classes. This approach helps optimize model performance, reduce bias, and enhance robustness. Additionally, hyperparameter optimization is performed using Optuna, ensuring that each algorithm operates at its optimal level. The results show that our model, using the Random Forest algorithm, achieves an accuracy of 99.80%, surpassing the performance of XGBoost and Deep Neural Networks (DNN). This makes our approach a true asset for intrusion detection methods in computer networks.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"22 ","pages":"Article 100738"},"PeriodicalIF":4.9,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145222733","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-10-17DOI: 10.1016/j.mlwa.2025.100761
Suresan Pareth
Ill-posed inverse problems frequently arise in scientific and medical imaging, where recovering stable and high-fidelity solutions from incomplete or noisy data remains a central challenge. Motivated by this need, we propose a novel hybrid solver framework, the Neural-Enhanced Two-Step Modified Newton–Lavrentiev Method (NE-TSMNLM), which integrates deep neural corrections into the classical Two-Step Modified Newton–Lavrentiev Method for solving nonlinear inverse problems. Unlike black-box neural operators, our design preserves the convergence structure of the classical iteration while embedding neural modules for adaptive correction, regularization, and convergence prediction.
We establish theoretical guarantees on stability and convergence: under mild assumptions, the NE-TSMNLM method inherits the convergence of the classical TSMNLM and improves the effective convergence rate to with . This demonstrates the acceleration effect due to neural corrections, which has been theoretically proven.
We validate the proposed framework on synthetic and medical inverse problems, including low-dose Computed Tomography (CT) reconstruction, where NE-TSMNLM achieves a 50% radiation dose reduction while maintaining structural fidelity. Initial implementations show promising results with slight degradation (e.g., 17.3% error increase) due to untrained modules and data scarcity. We identify clear pathways for improvement using Transformer-based modules, residual-aware training, and scalable synthetic data.
These results position NE-TSMNLM as a structure-preserving neural framework with rigorous mathematical guarantees, bridging classical regularization theory and deep learning for stable, efficient, and interpretable scientific machine learning.
{"title":"Neural-Enhanced Two-Step Modified Newton–Lavrentiev Method: A structure-preserving deep learning approach for ill-posed inverse problems","authors":"Suresan Pareth","doi":"10.1016/j.mlwa.2025.100761","DOIUrl":"10.1016/j.mlwa.2025.100761","url":null,"abstract":"<div><div>Ill-posed inverse problems frequently arise in scientific and medical imaging, where recovering stable and high-fidelity solutions from incomplete or noisy data remains a central challenge. Motivated by this need, we propose a novel hybrid solver framework, the <strong>Neural-Enhanced Two-Step Modified Newton–Lavrentiev Method (NE-TSMNLM)</strong>, which integrates deep neural corrections into the classical Two-Step Modified Newton–Lavrentiev Method for solving nonlinear inverse problems. Unlike black-box neural operators, our design preserves the convergence structure of the classical iteration while embedding neural modules for adaptive correction, regularization, and convergence prediction.</div><div>We establish theoretical guarantees on stability and convergence: under mild assumptions, the NE-TSMNLM method inherits the convergence of the classical TSMNLM and improves the effective convergence rate to <span><math><mrow><mover><mrow><mi>q</mi></mrow><mrow><mo>̃</mo></mrow></mover><mo>=</mo><msup><mrow><mi>q</mi></mrow><mrow><mn>1</mn><mo>+</mo><mi>β</mi></mrow></msup></mrow></math></span> with <span><math><mrow><mi>β</mi><mo>></mo><mn>0</mn></mrow></math></span>. This demonstrates the acceleration effect due to neural corrections, which has been theoretically proven.</div><div>We validate the proposed framework on synthetic and medical inverse problems, including low-dose Computed Tomography (CT) reconstruction, where NE-TSMNLM achieves a 50% radiation dose reduction while maintaining structural fidelity. Initial implementations show promising results with slight degradation (e.g., 17.3% error increase) due to untrained modules and data scarcity. We identify clear pathways for improvement using Transformer-based modules, residual-aware training, and scalable synthetic data.</div><div>These results position NE-TSMNLM as a structure-preserving neural framework with rigorous mathematical guarantees, bridging classical regularization theory and deep learning for stable, efficient, and interpretable scientific machine learning.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"22 ","pages":"Article 100761"},"PeriodicalIF":4.9,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145321181","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
With the advent of digital microscopy, the International Council for Standardization in Hematology recommends digital imaging and artificial intelligence (AI) algorithms for automatically classifying blood cells in peripheral blood smears to enhance diagnostic efficiency and accuracy. Nevertheless, while early AI studies have shown promising results in classifying white blood cells, the prediction process often remains unclear. Herein, we aimed to build a highly accurate model and visualize the basis of its predictions. The dataset comprised peripheral blood smear images of normal cells from individuals without infections, hematological disorders, or tumors, who were not undergoing any drug treatment at the time of blood collection. The images were obtained using a CellaVision DM96 analyzer at the Core Laboratory of Hospital Clínic de Barcelona. We used VGG16 and ResNet50 with transfer learning on ImageNet and applied the Grad-CAM method to visualize the image regions on which the model focused for classification. The model effectively recognized features, such as nuclear indentation and cytoplasmic color, which are crucial for classifying promyelocytes, myelocytes, and metamyelocytes. Traditionally, the basis of AI model predictions has been opaque, posing a challenge for medical applications. Our visualized classification basis clarifies the decision-making process of the model. These insights suggest that understanding these features can make the predictions of AI models more reliable and interpretable. Our findings improve diagnostic efficiency and suggest the potential of AI-based diagnostic support systems. Future research should validate this model’s performance using more extensive datasets and different cell types to enhance its reliability and practicality.
{"title":"Toward clinical reliability: Visualizing and interpreting ai-based classification in peripheral blood smear analysis","authors":"Hiroaki Iwata, Tsukie Shibayama, Miku Watanabe, Hisashi Shimohiro","doi":"10.1016/j.mlwa.2025.100780","DOIUrl":"10.1016/j.mlwa.2025.100780","url":null,"abstract":"<div><div>With the advent of digital microscopy, the International Council for Standardization in Hematology recommends digital imaging and artificial intelligence (AI) algorithms for automatically classifying blood cells in peripheral blood smears to enhance diagnostic efficiency and accuracy. Nevertheless, while early AI studies have shown promising results in classifying white blood cells, the prediction process often remains unclear. Herein, we aimed to build a highly accurate model and visualize the basis of its predictions. The dataset comprised peripheral blood smear images of normal cells from individuals without infections, hematological disorders, or tumors, who were not undergoing any drug treatment at the time of blood collection. The images were obtained using a CellaVision DM96 analyzer at the Core Laboratory of Hospital Clínic de Barcelona. We used VGG16 and ResNet50 with transfer learning on ImageNet and applied the Grad-CAM method to visualize the image regions on which the model focused for classification. The model effectively recognized features, such as nuclear indentation and cytoplasmic color, which are crucial for classifying promyelocytes, myelocytes, and metamyelocytes. Traditionally, the basis of AI model predictions has been opaque, posing a challenge for medical applications. Our visualized classification basis clarifies the decision-making process of the model. These insights suggest that understanding these features can make the predictions of AI models more reliable and interpretable. Our findings improve diagnostic efficiency and suggest the potential of AI-based diagnostic support systems. Future research should validate this model’s performance using more extensive datasets and different cell types to enhance its reliability and practicality.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"22 ","pages":"Article 100780"},"PeriodicalIF":4.9,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145466476","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-09-10DOI: 10.1016/j.mlwa.2025.100734
Hamid Mirzahossein, Soheil Rezashoar
Customer experience is crucial in the airline industry, as understanding passenger satisfaction helps airlines improve service quality. This study evaluates hyperparameter optimization and feature interpretability in machine learning models for predicting airline passenger satisfaction. Support Vector Machine (SVM) and Multilayer Perceptron (MLP) models were tested for binary classification, labeling passengers as ‘Satisfied’ or ‘Neutral or Dissatisfied’ using a Kaggle dataset with ∼104,000 training and ∼26,000 test records. Hyperparameter tuning used grid search with 10-fold cross-validation. For SVM, the optimal setup included the RBF kernel, C = 10, and gamma = ‘auto’, achieving a mean score of 0.9606. For MLP, the best configuration used no regularization, "he" initialization, ReLU activation, 30 epochs, batch size of 32, two hidden layers with 32 neurons each, and a learning rate of 0.001, yielding a mean score of 0.9556. Performance metrics included accuracy, precision, recall, and F1-Score, with SVM achieving a test accuracy of 0.96, precision of 0.97, and F1-Score of 0.95, slightly outperforming MLP by <1 %, though MLP was faster at 0.3 s versus SVM’s 18 s. Both models surpassed baseline models and prior studies, benefiting from robust preprocessing and a large dataset. Permutation importance analysis identified Type of Travel, Inflight Wi-Fi Service, Customer Type, and Online Boarding as key predictors, emphasizing passenger needs for digital connectivity and personalized services. These insights guide airlines to prioritize reliable Wi-Fi and efficient online boarding to enhance satisfaction, loyalty, and competitive positioning.
{"title":"Feature importance analysis of optimized machine learning modeling for predicting customers satisfaction at the United States Airlines","authors":"Hamid Mirzahossein, Soheil Rezashoar","doi":"10.1016/j.mlwa.2025.100734","DOIUrl":"10.1016/j.mlwa.2025.100734","url":null,"abstract":"<div><div>Customer experience is crucial in the airline industry, as understanding passenger satisfaction helps airlines improve service quality. This study evaluates hyperparameter optimization and feature interpretability in machine learning models for predicting airline passenger satisfaction. Support Vector Machine (SVM) and Multilayer Perceptron (MLP) models were tested for binary classification, labeling passengers as ‘Satisfied’ or ‘Neutral or Dissatisfied’ using a Kaggle dataset with ∼104,000 training and ∼26,000 test records. Hyperparameter tuning used grid search with 10-fold cross-validation. For SVM, the optimal setup included the RBF kernel, <em>C</em> = 10, and gamma = ‘auto’, achieving a mean score of 0.9606. For MLP, the best configuration used no regularization, \"he\" initialization, ReLU activation, 30 epochs, batch size of 32, two hidden layers with 32 neurons each, and a learning rate of 0.001, yielding a mean score of 0.9556. Performance metrics included accuracy, precision, recall, and F1-Score, with SVM achieving a test accuracy of 0.96, precision of 0.97, and F1-Score of 0.95, slightly outperforming MLP by <1 %, though MLP was faster at 0.3 s versus SVM’s 18 s. Both models surpassed baseline models and prior studies, benefiting from robust preprocessing and a large dataset. Permutation importance analysis identified Type of Travel, Inflight Wi-Fi Service, Customer Type, and Online Boarding as key predictors, emphasizing passenger needs for digital connectivity and personalized services. These insights guide airlines to prioritize reliable Wi-Fi and efficient online boarding to enhance satisfaction, loyalty, and competitive positioning.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"22 ","pages":"Article 100734"},"PeriodicalIF":4.9,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145108030","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-11-07DOI: 10.1016/j.mlwa.2025.100792
Mohammad Amin Amiri , Saeid Afshari , Ali Soltani
Road traffic injuries continue to pose a significant public health challenge in Australia, with pedestrians representing one of the most vulnerable road user groups. Accurate prediction of injury severity, particularly fatal outcomes, is essential for improving road safety interventions and resource allocation. This study applies advanced machine learning techniques to predict pedestrian crash severity using national hospitalization and mortality data collected from 2011 to 2021. The analysis focuses on addressing class imbalance, a common issue in injury data by evaluating the impact of several data balancing methods, including SMOTE, ADASYN, Random Oversampling (ROS), and Threshold Moving. We implement and compare four supervised learning algorithms: Logistic Regression, Support Vector Machine (SVM), Decision Tree, and XGBoost. Model performance is assessed using F1-score and macro-accuracy, with a focus on the minority (fatality) class. Results show that XGBoost combined with Threshold Moving achieves the highest performance, yielding an F1-score of 72% for fatality classification and a macro-accuracy of 84%. Additionally, feature importance analysis using SHAP values reveals age, gender, road user type, and crash location as key predictors of injury severity. The study highlights the critical role of data balancing strategies in enhancing predictive accuracy for rare but high-impact outcomes. These findings provide actionable insights for transport authorities and policymakers seeking to develop data-driven, targeted safety measures to protect pedestrians and reduce the severity of crash outcomes.
{"title":"Machine learning approaches to traffic accident severity prediction: Addressing class imbalance","authors":"Mohammad Amin Amiri , Saeid Afshari , Ali Soltani","doi":"10.1016/j.mlwa.2025.100792","DOIUrl":"10.1016/j.mlwa.2025.100792","url":null,"abstract":"<div><div>Road traffic injuries continue to pose a significant public health challenge in Australia, with pedestrians representing one of the most vulnerable road user groups. Accurate prediction of injury severity, particularly fatal outcomes, is essential for improving road safety interventions and resource allocation. This study applies advanced machine learning techniques to predict pedestrian crash severity using national hospitalization and mortality data collected from 2011 to 2021. The analysis focuses on addressing class imbalance, a common issue in injury data by evaluating the impact of several data balancing methods, including SMOTE, ADASYN, Random Oversampling (ROS), and Threshold Moving. We implement and compare four supervised learning algorithms: Logistic Regression, Support Vector Machine (SVM), Decision Tree, and XGBoost. Model performance is assessed using F1-score and macro-accuracy, with a focus on the minority (fatality) class. Results show that XGBoost combined with Threshold Moving achieves the highest performance, yielding an F1-score of 72% for fatality classification and a macro-accuracy of 84%. Additionally, feature importance analysis using SHAP values reveals age, gender, road user type, and crash location as key predictors of injury severity. The study highlights the critical role of data balancing strategies in enhancing predictive accuracy for rare but high-impact outcomes. These findings provide actionable insights for transport authorities and policymakers seeking to develop data-driven, targeted safety measures to protect pedestrians and reduce the severity of crash outcomes.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"22 ","pages":"Article 100792"},"PeriodicalIF":4.9,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145528029","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-11-06DOI: 10.1016/j.mlwa.2025.100786
Movaffaq Kateb , Sahar Safarian
This work demonstrates that modern tree‑based models can effectively model complex, temperature-dependent mechanical responses, including highly nonlinear and even non-monotonic trends, in austenitic stainless steel and highlights limitations of composition‑only empirical models. To ensure robust model evaluation, we employed multiple validation strategies including repeated random train and test partitions and leave-one-out cross-validation. While one might assume that steel grade is fully captured by its composition, local assessments within narrower compositional ranges reveal different feature importance rankings than those observed in the full dataset. Grade-specific (AISI 304, 316, 321 and 347) feature importance analysis offered deeper insights into local alloy behavior and demonstrated the advantage of disaggregated modeling in avoiding misleading conclusions. Clustering and SHAP analyses further revealed a temperature-sensitive role of nitrogen, which strengthens the alloy through interstitial and fine precipitate mechanisms at lower temperatures but loses effectiveness at elevated temperatures due to precipitate coarsening. This highlights how data-driven methods can uncover metallurgically consistent, temperature-dependent strengthening behaviors not captured by simpler models. Our results confirm that temperature governs the mechanical performance of austenitic stainless steels, with other features contributing marginally, particularly for UTS. Additionally, the model achieved a notably high score for elongation, highlighting the critical role of testing temperature in addressing the long-standing challenge of poor elongation predictions in composition-only or composition-processing models. This suggests that low accuracy in previous studies is more likely due to dataset limitations rather than shortcomings of tree-based models.
{"title":"Machine learning-driven predictive modeling of temperature-dependent mechanical properties in austenitic stainless steels","authors":"Movaffaq Kateb , Sahar Safarian","doi":"10.1016/j.mlwa.2025.100786","DOIUrl":"10.1016/j.mlwa.2025.100786","url":null,"abstract":"<div><div>This work demonstrates that modern tree‑based models can effectively model complex, temperature-dependent mechanical responses, including highly nonlinear and even non-monotonic trends, in austenitic stainless steel and highlights limitations of composition‑only empirical models. To ensure robust model evaluation, we employed multiple validation strategies including repeated random train and test partitions and leave-one-out cross-validation. While one might assume that steel grade is fully captured by its composition, local assessments within narrower compositional ranges reveal different feature importance rankings than those observed in the full dataset. Grade-specific (AISI 304, 316, 321 and 347) feature importance analysis offered deeper insights into local alloy behavior and demonstrated the advantage of disaggregated modeling in avoiding misleading conclusions. Clustering and SHAP analyses further revealed a temperature-sensitive role of nitrogen, which strengthens the alloy through interstitial and fine precipitate mechanisms at lower temperatures but loses effectiveness at elevated temperatures due to precipitate coarsening. This highlights how data-driven methods can uncover metallurgically consistent, temperature-dependent strengthening behaviors not captured by simpler models. Our results confirm that temperature governs the mechanical performance of austenitic stainless steels, with other features contributing marginally, particularly for UTS. Additionally, the model achieved a notably high score for elongation, highlighting the critical role of testing temperature in addressing the long-standing challenge of poor elongation predictions in composition-only or composition-processing models. This suggests that low accuracy in previous studies is more likely due to dataset limitations rather than shortcomings of tree-based models.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"22 ","pages":"Article 100786"},"PeriodicalIF":4.9,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145528530","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-11-12DOI: 10.1016/j.mlwa.2025.100794
Sara Khan , Mehmed Yüksel , Frank Kirchner
Wear and tear detection in fleet and shared vehicle systems is a critical challenge, particularly in rental and car-sharing services, where minor damage, such as dents, scratches, and underbody impacts, often goes unnoticed or is detected too late. Currently, manual inspection methods are the default approach, but are labour-intensive and prone to human error. In contrast, state-of-the-art image-based methods are less reliable when the vehicle is moving, and they cannot effectively capture underbody damage due to limited visual access and spatial coverage. This work introduces a novel multi-modal architecture based on anomaly detection to address these issues. Sensors such as Inertial Measurement Units (IMUs) and microphones are integrated into a compact device mounted on the vehicle’s windshield. This approach supports real-time damage detection while avoiding the need for highly resource-intensive sensors. We developed multiple variants of multi-modal autoencoder-based architectures and evaluated them against unimodal and state-of-the-art methods. Our multi-modal ensemble model with pooling achieved the highest performance, with a Receiver Operating Characteristic-Area Under Curve (ROC-AUC) of 92%, demonstrating its effectiveness in real-world applications. This approach can also be extended to other applications, such as improving automotive safety. It can integrate with airbag systems for efficient deployment and help autonomous vehicles by complementing other sensors in collision detection.
{"title":"Robust anomaly detection through multi-modal autoencoder fusion for small vehicle damage detection","authors":"Sara Khan , Mehmed Yüksel , Frank Kirchner","doi":"10.1016/j.mlwa.2025.100794","DOIUrl":"10.1016/j.mlwa.2025.100794","url":null,"abstract":"<div><div>Wear and tear detection in fleet and shared vehicle systems is a critical challenge, particularly in rental and car-sharing services, where minor damage, such as dents, scratches, and underbody impacts, often goes unnoticed or is detected too late. Currently, manual inspection methods are the default approach, but are labour-intensive and prone to human error. In contrast, state-of-the-art image-based methods are less reliable when the vehicle is moving, and they cannot effectively capture underbody damage due to limited visual access and spatial coverage. This work introduces a novel multi-modal architecture based on anomaly detection to address these issues. Sensors such as Inertial Measurement Units (IMUs) and microphones are integrated into a compact device mounted on the vehicle’s windshield. This approach supports real-time damage detection while avoiding the need for highly resource-intensive sensors. We developed multiple variants of multi-modal autoencoder-based architectures and evaluated them against unimodal and state-of-the-art methods. Our multi-modal ensemble model with pooling achieved the highest performance, with a Receiver Operating Characteristic-Area Under Curve (ROC-AUC) of 92%, demonstrating its effectiveness in real-world applications. This approach can also be extended to other applications, such as improving automotive safety. It can integrate with airbag systems for efficient deployment and help autonomous vehicles by complementing other sensors in collision detection.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"22 ","pages":"Article 100794"},"PeriodicalIF":4.9,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145528531","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-09-04DOI: 10.1016/j.mlwa.2025.100712
Yinsheng Zhang, Mingming He, Haiyan Wang
Logistic regression is a simple yet widely used classification model in spectroscopic profiling analysis. Considering the model’s output represents a probability, this paper will investigate its latent distribution assumption, i.e., its inner linear regressor unit follows a standard logistic distribution. An empirical study on five spectroscopic profiling open datasets, i.e., wine, coffee, olive oil, cheese, and milk powder, was conducted to verify this latent distribution assertion. This paper measured the GoF (Goodness of Fit) of each dataset’s latent variable from three aspects, i.e., curve fitting, P–P and Q–Q plots, and K–S test. After hyper-parameter optimization and proper training, the latent variable, as a weighted sum of the original features, has demonstrated a high level of GoF on all the five datasets. This study verifies the suitability of logistic regression in spectroscopic profiling analysis and answers why the model output can be interpreted as a conditional probability.
{"title":"On the latent distribution of logistic regression — An empirical study on spectroscopic profiling datasets","authors":"Yinsheng Zhang, Mingming He, Haiyan Wang","doi":"10.1016/j.mlwa.2025.100712","DOIUrl":"10.1016/j.mlwa.2025.100712","url":null,"abstract":"<div><div>Logistic regression is a simple yet widely used classification model in spectroscopic profiling analysis. Considering the model’s output represents a probability, this paper will investigate its latent distribution assumption, i.e., its inner linear regressor unit follows a standard logistic distribution. An empirical study on five spectroscopic profiling open datasets, i.e., wine, coffee, olive oil, cheese, and milk powder, was conducted to verify this latent distribution assertion. This paper measured the GoF (Goodness of Fit) of each dataset’s latent variable from three aspects, i.e., curve fitting, P–P and Q–Q plots, and K–S test. After hyper-parameter optimization and proper training, the latent variable, as a weighted sum of the original features, has demonstrated a high level of GoF on all the five datasets. This study verifies the suitability of logistic regression in spectroscopic profiling analysis and answers why the model output can be interpreted as a conditional probability.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"22 ","pages":"Article 100712"},"PeriodicalIF":4.9,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145050591","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-10-17DOI: 10.1016/j.mlwa.2025.100759
Giacomo Ibba , Rumyana Neykova , Marco Ortu , Roberto Tonelli , Steve Counsell , Giuseppe Destefanis
This paper introduces a methodology for software vulnerability detection that combines structural and semantic analysis through software metrics and topic modelling. We evaluate the approach using smart contracts as a case study, focusing on their structural properties and the presence of known security vulnerabilities. We identify the most relevant metrics for vulnerability detection, evaluate multiple machine learning classifiers for both binary and multi-label classification, and improve classification performance by integrating topic modelling techniques.
Our analysis shows that metrics such as cyclomatic complexity, nesting depth, and function calls are strongly associated with vulnerability presence. Using these metrics, the Random Forest classifier achieved strong performance in binary classification (AUC: 0.982, accuracy: 0.977, F1-score: 0.808) and multi-label classification (AUC: 0.951, accuracy: 0.729, F1-score: 0.839). The addition of topic modelling using Non-Negative Matrix Factorisation further improved results, increasing the F1-score to 0.881. The evaluation is conducted on Ethereum smart contracts written in Solidity.
{"title":"A machine learning approach to vulnerability detection combining software metrics and topic modelling: Evidence from smart contracts","authors":"Giacomo Ibba , Rumyana Neykova , Marco Ortu , Roberto Tonelli , Steve Counsell , Giuseppe Destefanis","doi":"10.1016/j.mlwa.2025.100759","DOIUrl":"10.1016/j.mlwa.2025.100759","url":null,"abstract":"<div><div>This paper introduces a methodology for software vulnerability detection that combines structural and semantic analysis through software metrics and topic modelling. We evaluate the approach using smart contracts as a case study, focusing on their structural properties and the presence of known security vulnerabilities. We identify the most relevant metrics for vulnerability detection, evaluate multiple machine learning classifiers for both binary and multi-label classification, and improve classification performance by integrating topic modelling techniques.</div><div>Our analysis shows that metrics such as cyclomatic complexity, nesting depth, and function calls are strongly associated with vulnerability presence. Using these metrics, the Random Forest classifier achieved strong performance in binary classification (AUC: 0.982, accuracy: 0.977, F1-score: 0.808) and multi-label classification (AUC: 0.951, accuracy: 0.729, F1-score: 0.839). The addition of topic modelling using Non-Negative Matrix Factorisation further improved results, increasing the F1-score to 0.881. The evaluation is conducted on Ethereum smart contracts written in Solidity.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"22 ","pages":"Article 100759"},"PeriodicalIF":4.9,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145363475","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-12-01Epub Date: 2025-11-01DOI: 10.1016/j.mlwa.2025.100773
Dorianis M. Perez, Bryan E. Kaiser, Ismael Boureima
As reasoning capabilities of large language models (LLMs) continue to advance, they are being integrated into increasingly complex scientific workflows, with the goal of developing agents capable of generating evidence-based explanations and testing hypotheses and theories. However, despite their rapid progress, most existing evaluations of LLM reasoning focus on accuracy or consistency rather than on uncertainty quantification (UQ), which is essential for evidence-based reasoning because it quantifies the degree of trustworthiness of evidence-based explanations. Current approaches to LLM uncertainty remain fragmented, often lacking standardized benchmarks that test models under varying task complexities. To address this gap, we introduce the first benchmark suite designed to evaluate UQ by LLM-based agents and tools. The benchmark targets one of the most fundamental UQ problem: estimating whether one quantity is probably larger than another under uncertainty. It includes two progressively complex tasks: a simple inequality test, where models judge whether one of two sets of samples is “larger,” “smaller,” or “uncertain” with 95% confidence, and a complex inequality test, where models assess interventional probabilities requiring multiple intermediate calculations. We found that reasoning models are generally capable of UQ (scores ) in the simple inequality case but do not score appreciably better than random guessing (scores ) for the complex inequality case if the UQ method and intermediate steps are not provided in the prompt. Our implementation is available at https://github.com/bekaiser-LANL/tether.
{"title":"Uncertainty quantification by large language models","authors":"Dorianis M. Perez, Bryan E. Kaiser, Ismael Boureima","doi":"10.1016/j.mlwa.2025.100773","DOIUrl":"10.1016/j.mlwa.2025.100773","url":null,"abstract":"<div><div>As reasoning capabilities of large language models (LLMs) continue to advance, they are being integrated into increasingly complex scientific workflows, with the goal of developing agents capable of generating evidence-based explanations and testing hypotheses and theories. However, despite their rapid progress, most existing evaluations of LLM reasoning focus on accuracy or consistency rather than on uncertainty quantification (UQ), which is essential for evidence-based reasoning because it quantifies the degree of trustworthiness of evidence-based explanations. Current approaches to LLM uncertainty remain fragmented, often lacking standardized benchmarks that test models under varying task complexities. To address this gap, we introduce the first benchmark suite designed to evaluate UQ by LLM-based agents and tools. The benchmark targets one of the most fundamental UQ problem: estimating whether one quantity is probably larger than another under uncertainty. It includes two progressively complex tasks: a simple inequality test, where models judge whether one of two sets of samples is “larger,” “smaller,” or “uncertain” with 95% confidence, and a complex inequality test, where models assess interventional probabilities requiring multiple intermediate calculations. We found that reasoning models are generally capable of UQ (scores <span><math><mrow><mo>≳</mo><mn>70</mn><mtext>%</mtext></mrow></math></span>) in the simple inequality case but do not score appreciably better than random guessing (scores <span><math><mrow><mo>∼</mo><mn>33</mn><mtext>%</mtext></mrow></math></span>) for the complex inequality case if the UQ method and intermediate steps are not provided in the prompt. Our implementation is available at <span><span>https://github.com/bekaiser-LANL/tether</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"22 ","pages":"Article 100773"},"PeriodicalIF":4.9,"publicationDate":"2025-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145466471","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号