Convolutional Neural Network-Based ECG Signal Classification Model: A Study on Addressing Class Imbalance and Enhancing Model Interpretability DOI Open Access
Guanjun Wang, Shuwen Zheng, Xuejun Yang

et al.

Published: May 20, 2024

Convolutional Neural Networks (CNNs) are often criticized for their lack of transparency, acting as 'black boxes' in decision-making, a challenge compounded by class imbalance ECG datasets, which limits clinical application. This study introduces CNN-based signal classification model that enhances interpretability and addresses through the Synthetic Minority Over-sampling Technique (SMOTE). The also integrates Uniform Manifold Approximation Projection (UMAP) dimensionality reduction SHAP value analysis, facilitating visualization decision boundaries assessment feature contributions. Our evaluation using MIT-BIH Arrhythmia Database highlights model's high performance, with accuracy precision nearing 1.00 Normal (NOR), Left Bundle Branch Block (LBBB), Right (RBBB), Ventricular Premature Beat (PVC) six-class task. In ten-class task, demonstrated robustness, particularly an 0.9846, 0.9783, recall 0.9736, F1 score 0.9760 Pacemaker Fusion (PFHB), supported AUC 0.9999 AP 0.9885. These results underscore efficacy cardiac rhythm recognition resilience to imbalance. Future research will explore sophisticated architectures extraction methods enhance generalization applicability early heart disease diagnosis personalized treatment.

Language: Английский

Machine learning-driven prediction of biochar adsorption capacity for effective removal of Congo red dye DOI Creative Commons
Shubham Yadav,

P. K. Rajput,

P. Balasubramanian

et al.

Carbon Research, Journal Year: 2025, Volume and Issue: 4(1)

Published: Jan. 22, 2025

Abstract Congo red, a widely utilized dye in the textile industry, presents significant threat to living organisms due its carcinogenic properties and non-biodegradable nature. This study proposes data-driven machine-learning approach optimize biochar characteristics environmental conditions maximize adsorption capacity of for removal red dye. Therefore, six machine learning models were trained tested on dataset containing eleven input parameters (related conditions) capacity. The evaluated using performance metrics such as R-squared ( R 2 ), Mean Squared Error (MSE), Root (RMSE). With highest (0.9785) lowest RMSE (0.1357), Random Forest Regression (RF) outperformed other models. DT XGB also performed well, achieving slightly lower values 0.9741 0.9577, respectively. LR model worst, with (0.4575) (0.6821). Moreover, reliability these was validated 10-fold cross-validation method. RF once again best an value 0.9762. Feature analysis revealed that initial concentration relative dosage C 0 specific surface area BET pore volume PV ) are most factors affecting biochar, while carbon content oxygen nitrogen molar ratio [ (O + N)/C ], diameter D had minimal impact. research demonstrates can accurately predict biochar’s contaminant capacity, enhancing wastewater treatment promoting efficient, cost-effective management. Graphical

Language: Английский

Citations

1
Investigating the Effect of Pore Size Distribution on the Sorption Types and the Adsorption-Deformation Characteristics of Porous Continua: The Case of Adsorption on Carbonaceous Materials DOI Creative Commons
Grigorios L. Kyriakopoulos,

Konstantinos Tsimnadis,

Ioannis Sebos

et al.

Crystals, Journal Year: 2024, Volume and Issue: 14(8), P. 742 - 742

Published: Aug. 20, 2024

In the chemical industry and in manufacturing sector, adsorption properties of porous materials have been proven to be great interest for removal impurities from liquid gas media. While it is acknowledged that significant progress literature production developed this field, there studies failed further advance our knowledge generating a better understanding prevailing sorption types dominant processes. Therefore, review study has focused on materials, their properties, investigating at either solid–gas solid–liquid interfaces, underscoring both characterization correlation between porosity capacity, as well emergent interactions adsorbent adsorbate molecules, including mechanisms, kinetic thermodynamic information conveyed.

Language: Английский

Citations

7
A Systematic Literature Review of the Latest Advancements in XAI DOI Creative Commons
Zaid M. Altukhi, Sojen Pradhan, Nasser Aljohani

et al.

Technologies, Journal Year: 2025, Volume and Issue: 13(3), P. 93 - 93

Published: March 1, 2025

This systematic review details recent advancements in the field of Explainable Artificial Intelligence (XAI) from 2014 to 2024. XAI utilises a wide range frameworks, techniques, and methods used interpret machine learning (ML) black-box models. We aim understand technical future directions. followed PRISMA methodology selected 30 relevant publications three main databases: IEEE Xplore, ACM, ScienceDirect. Through comprehensive thematic analysis, we categorised research into topics: ‘model developments’, ‘evaluation metrics methods’, ‘user-centred system design’. Our results uncover ‘What’, ‘How’, ‘Why’ these were developed. found that 13 papers focused on model developments, 8 studies evaluation metrics, 12 user-centred design. Moreover, it was aimed bridge gap between outputs user understanding.

Language: Английский

Citations

0
Application of Machine Learning for Predicting the Incubation Period of Water Droplet Erosion in Metals DOI

Khaled AlHammad,

Mamoun Medraj, Moussa Tembely

et al.

Research Square (Research Square), Journal Year: 2025, Volume and Issue: unknown

Published: April 16, 2025

Abstract Water droplet erosion (WDE) is a critical phenomenon that leads to material degradation in many engineering applications, particularly power generation and aerospace industry. Accurate prediction of the incubation period essential for optimizing selection maintenance strategies. Traditional empirical models, while helpful, often lack predictive accuracy due their reliance on numerous parameters with limited physical interpretation. In this study, machine learning (ML) approach was developed predict different materials. A range ML models—including linear regression (LR), decision tree regressor (DT), random forest (RF), gradient boosting (GBR), artificial neural networks (ANN)—was employed capture complex relationships between properties conditions. Despite hyperparameter optimization using techniques such as grid search, no substantial improvement model predictions observed. Data transformation methods—logarithmic, Yeo-Johnson, Box-Cox transformations—were applied enhance performance. dataset derived from experimental measurements five alloys used train validate models. The results indicate models significantly outperform conventional approaches. Notably, LR achieved an R² (coefficient determination) over 90% low mean absolute error (MAE), ANN Yeo-Johnson attained above 85% correspondingly MAE. Additionally, feature impact importance analyses provided insights into key factors influencing duration period, further validating robustness This study offers robust tool predicting WDE, broad applicability design across various industries.

Language: Английский

Citations

0
Optimizing Lithium-Ion Battery Performance: Integrating Machine Learning and Explainable AI for Enhanced Energy Management DOI Open Access
Saadin Oyucu, Betül Ersöz, Şeref Sağıroğlu

et al.

Sustainability, Journal Year: 2024, Volume and Issue: 16(11), P. 4755 - 4755

Published: June 3, 2024

Managing the capacity of lithium-ion batteries (LiBs) accurately, particularly in large-scale applications, enhances cost-effectiveness energy storage systems. Less frequent replacement or maintenance LiBs results cost savings long term. Therefore, this study, AdaBoost, gradient boosting, XGBoost, LightGBM, CatBoost, and ensemble learning models were employed to predict discharge LiBs. The prediction performances each model compared based on mean absolute error (MAE), squared (MSE), R-squared values. research findings reveal that LightGBM exhibited lowest MAE (0.103) MSE (0.019) values highest (0.887) value, thus demonstrating strongest correlation predictions. Gradient boosting XGBoost showed similar performance levels but ranked just below LightGBM. competitive indicates combining multiple could lead an overall improvement. Furthermore, study incorporates analysis key features affecting predictions using SHAP (Shapley additive explanations) within framework explainable artificial intelligence (XAI). This evaluates impact such as temperature, cycle index, voltage, current predictions, revealing a significant effect temperature capacity. emphasize potential machine LiB management XAI demonstrate how these technologies play strategic role optimizing

Language: Английский

Citations

3
Convolutional Neural Network-Based ECG Signal Classification Model: A Study on Addressing Class Imbalance and Enhancing Model Interpretability DOI Open Access
Guanjun Wang, Shuwen Zheng, Xuejun Yang

et al.

Published: May 20, 2024

Convolutional Neural Networks (CNNs) are often criticized for their lack of transparency, acting as 'black boxes' in decision-making, a challenge compounded by class imbalance ECG datasets, which limits clinical application. This study introduces CNN-based signal classification model that enhances interpretability and addresses through the Synthetic Minority Over-sampling Technique (SMOTE). The also integrates Uniform Manifold Approximation Projection (UMAP) dimensionality reduction SHAP value analysis, facilitating visualization decision boundaries assessment feature contributions. Our evaluation using MIT-BIH Arrhythmia Database highlights model's high performance, with accuracy precision nearing 1.00 Normal (NOR), Left Bundle Branch Block (LBBB), Right (RBBB), Ventricular Premature Beat (PVC) six-class task. In ten-class task, demonstrated robustness, particularly an 0.9846, 0.9783, recall 0.9736, F1 score 0.9760 Pacemaker Fusion (PFHB), supported AUC 0.9999 AP 0.9885. These results underscore efficacy cardiac rhythm recognition resilience to imbalance. Future research will explore sophisticated architectures extraction methods enhance generalization applicability early heart disease diagnosis personalized treatment.

Language: Английский

Citations

1