Machine learning with applications最新文献_第2页

Apply a deep learning hybrid model optimized by an Improved Chimp Optimization Algorithm in PM2.5 prediction

Machine learning with applications

Pub Date : 2025-01-28 DOI: 10.1016/j.mlwa.2025.100624

Ming Wei, Xiaopeng Du

PM_2.5 pollution in the atmosphere not only contaminates the environment but also seriously affects human health. Therefore, studying how to accurately predict future PM_2.5 concentrations holds significant importance and practical value. This paper innovatively

P M_{2.5}

proposes a high-accuracy prediction model: RF-ICHOA-CNN-LSTM-Attention. First, the Random Forest (RF) model is utilized to evaluate the importance of air pollution and meteorological features and select more suitable input features. Subsequently, a one-dimensional convolutional neural network (1DCNN) with efficient feature extraction capability is used to extract dynamic features from sequences. The extracted feature vector sequences are then fed into a Long Short-Term Memory Network (LSTM). After the LSTM, an Attention Mechanism is incorporated to assign different weights to the input features, emphasizing the role of the important features. Additionally, the Improved Chimp Optimization Algorithm (IChOA) is employed to optimize the number of neurons in the two hidden layers of LSTM, the learning rate, and the number of training epochs. The experimental results on 12 test functions demonstrate that the optimization performance of IChOA is better than that of ChOA and the representative swarm optimization algorithms used for comparison. In the case of PM_2.5 predictions in Yining and Beijing, experimental results show that the proposed model achieved the best performance in terms of RMSE, MAE, and R

^{2}

This indicates its excellent prediction accuracy and generalization capability, Thus proving its effectiveness in predicting PM_2.5 concentration in the real world.

{"title":"Apply a deep learning hybrid model optimized by an Improved Chimp Optimization Algorithm in PM2.5 prediction","authors":"Ming Wei, Xiaopeng Du","doi":"10.1016/j.mlwa.2025.100624","DOIUrl":"10.1016/j.mlwa.2025.100624","url":null,"abstract":"<div><div>PM<sub>2.5</sub> pollution in the atmosphere not only contaminates the environment but also seriously affects human health. Therefore, studying how to accurately predict future PM<sub>2.5</sub> concentrations holds significant importance and practical value. This paper innovatively <span><math><mrow><mi>P</mi><msub><mrow><mi>M</mi></mrow><mrow><mn>2</mn><mo>.</mo><mn>5</mn></mrow></msub></mrow></math></span>proposes a high-accuracy prediction model: RF-ICHOA-CNN-LSTM-Attention. First, the Random Forest (RF) model is utilized to evaluate the importance of air pollution and meteorological features and select more suitable input features. Subsequently, a one-dimensional convolutional neural network (1DCNN) with efficient feature extraction capability is used to extract dynamic features from sequences. The extracted feature vector sequences are then fed into a Long Short-Term Memory Network (LSTM). After the LSTM, an Attention Mechanism is incorporated to assign different weights to the input features, emphasizing the role of the important features. Additionally, the Improved Chimp Optimization Algorithm (IChOA) is employed to optimize the number of neurons in the two hidden layers of LSTM, the learning rate, and the number of training epochs. The experimental results on 12 test functions demonstrate that the optimization performance of IChOA is better than that of ChOA and the representative swarm optimization algorithms used for comparison. In the case of PM<sub>2.5</sub> predictions in Yining and Beijing, experimental results show that the proposed model achieved the best performance in terms of RMSE, MAE, and R<span><math><msup><mrow></mrow><mrow><mn>2</mn></mrow></msup></math></span> This indicates its excellent prediction accuracy and generalization capability, Thus proving its effectiveness in predicting PM<sub>2.5</sub> concentration in the real world.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"19 ","pages":"Article 100624"},"PeriodicalIF":0.0,"publicationDate":"2025-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143170235","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Ensembles of deep one-class classifiers for multi-class image classification

Machine learning with applications

Pub Date : 2025-01-22 DOI: 10.1016/j.mlwa.2025.100621

Alexander Novotny , George Bebis , Alireza Tavakkoli , Mircea Nicolescu

Traditional methods for multi-class classification (MCC) involve using a monolithic feature extractor and classifier trained on data from all the classes simultaneously. These methods are dependent on the number and types of classes and are therefore rigid against changes to the class structure. For instance, if the number of classes needs to be modified or new training data becomes available, retraining would be required for optimum classification performance. Moreover, these classifiers can become biased toward classes with a large data imbalance. An alternative, more attractive framework is to consider an ensemble of one-class classifiers (EOCC) where each one-class classifier (OCC) is trained with data from a single class only, without using any information from the other classes. Although this framework has not yet systematically matched or surpassed the performance of traditional MCC approaches, it deserves further investigation for several reasons. First, it provides a more flexible framework for handling changes in class structure compared to the traditional MCC approach. Second, it is less biased toward classes with large data imbalances compared to the multi-class classification approach. Finally, each OCC can be separately optimized depending on the characteristics of the class it represents. In this paper, we have performed extensive experiments to evaluate EOCC for MCC using traditional OCCs based on Principal Component Analysis (PCA) and Auto-encoders (AE) as well as newly proposed OCCs based on Generative Adversarial Networks (GANs). Moreover, we have compared the performance of EOCC with traditional multi-class DL classifiers including VGG-19, Resnet and EfficientNet. Two different datasets were used in our experiments: (i) a subset from the Plant Village dataset plant disease dataset with high variance in the number of classes and amount of data in each class, and (ii) an Alzheimer’s disease dataset with low amounts of data and a large imbalance in data between classes. Our results show that the GAN-based EOCC outperform previous EOCC approaches and improve the performance gap with traditional MCC approaches.

{"title":"Ensembles of deep one-class classifiers for multi-class image classification","authors":"Alexander Novotny , George Bebis , Alireza Tavakkoli , Mircea Nicolescu","doi":"10.1016/j.mlwa.2025.100621","DOIUrl":"10.1016/j.mlwa.2025.100621","url":null,"abstract":"<div><div>Traditional methods for multi-class classification (MCC) involve using a monolithic feature extractor and classifier trained on data from all the classes simultaneously. These methods are dependent on the number and types of classes and are therefore rigid against changes to the class structure. For instance, if the number of classes needs to be modified or new training data becomes available, retraining would be required for optimum classification performance. Moreover, these classifiers can become biased toward classes with a large data imbalance. An alternative, more attractive framework is to consider an ensemble of one-class classifiers (EOCC) where each one-class classifier (OCC) is trained with data from a single class only, without using any information from the other classes. Although this framework has not yet systematically matched or surpassed the performance of traditional MCC approaches, it deserves further investigation for several reasons. First, it provides a more flexible framework for handling changes in class structure compared to the traditional MCC approach. Second, it is less biased toward classes with large data imbalances compared to the multi-class classification approach. Finally, each OCC can be separately optimized depending on the characteristics of the class it represents. In this paper, we have performed extensive experiments to evaluate EOCC for MCC using traditional OCCs based on Principal Component Analysis (PCA) and Auto-encoders (AE) as well as newly proposed OCCs based on Generative Adversarial Networks (GANs). Moreover, we have compared the performance of EOCC with traditional multi-class DL classifiers including VGG-19, Resnet and EfficientNet. Two different datasets were used in our experiments: (i) a subset from the Plant Village dataset plant disease dataset with high variance in the number of classes and amount of data in each class, and (ii) an Alzheimer’s disease dataset with low amounts of data and a large imbalance in data between classes. Our results show that the GAN-based EOCC outperform previous EOCC approaches and improve the performance gap with traditional MCC approaches.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"19 ","pages":"Article 100621"},"PeriodicalIF":0.0,"publicationDate":"2025-01-22","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143170233","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Safety analysis in the era of large language models: A case study of STPA using ChatGPT

Machine learning with applications

Pub Date : 2025-01-20 DOI: 10.1016/j.mlwa.2025.100622

Yi Qi , Xingyu Zhao , Siddartha Khastgir , Xiaowei Huang

Can safety analysis leverage Large Language Models (LLMs)? This study examines the application of Systems Theoretic Process Analysis (STPA) to Automatic Emergency Brake (AEB) and Electricity Demand Side Management (DSM) systems, utilising Chat Generative Pre-Trained Transformer (ChatGPT). We investigate the impact of collaboration schemes, input semantic complexity, and prompt engineering on STPA results. Comparative results indicate that using ChatGPT without human intervention may be inadequate due to reliability issues. However, with careful design, it has the potential to outperform human experts. No statistically significant differences were observed when varying the input semantic complexity or using domain-agnostic prompt guidelines. While STPA-specific prompt engineering produced statistically significant and more pertinent results, ChatGPT generally yielded more conservative and less comprehensive outcomes. We also identify future challenges, such as concerns regarding the trustworthiness of LLMs and the need for standardisation and regulation in this field. All experimental data are publicly accessible.

{"title":"Safety analysis in the era of large language models: A case study of STPA using ChatGPT","authors":"Yi Qi , Xingyu Zhao , Siddartha Khastgir , Xiaowei Huang","doi":"10.1016/j.mlwa.2025.100622","DOIUrl":"10.1016/j.mlwa.2025.100622","url":null,"abstract":"<div><div>Can safety analysis leverage Large Language Models (LLMs)? This study examines the application of Systems Theoretic Process Analysis (STPA) to Automatic Emergency Brake (AEB) and Electricity Demand Side Management (DSM) systems, utilising Chat Generative Pre-Trained Transformer (ChatGPT). We investigate the impact of collaboration schemes, input semantic complexity, and prompt engineering on STPA results. Comparative results indicate that using ChatGPT without human intervention may be inadequate due to reliability issues. However, with careful design, it has the potential to outperform human experts. No statistically significant differences were observed when varying the input semantic complexity or using domain-agnostic prompt guidelines. While STPA-specific prompt engineering produced statistically significant and more pertinent results, ChatGPT generally yielded more conservative and less comprehensive outcomes. We also identify future challenges, such as concerns regarding the trustworthiness of LLMs and the need for standardisation and regulation in this field. All experimental data are publicly accessible.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"19 ","pages":"Article 100622"},"PeriodicalIF":0.0,"publicationDate":"2025-01-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143170234","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Machine learning techniques and multi-objective programming to select the best suppliers and determine the orders

Machine learning with applications

Pub Date : 2025-01-17 DOI: 10.1016/j.mlwa.2025.100623

Asma ul Husna, Saman Hassanzadeh Amin, Ahmad Ghasempoor

Selection of appropriate suppliers and allocation the orders among them have become the two key strategic decisions regarding purchasing. In this study, a two-phase integrated approach is proposed for solving supplier selection and order allocation problems. Phase 1 contains four techniques from statistics and Machine Learning (ML), including Auto-Regressive Integrated Moving Average, Random Forest, Gradient Boosting Regression, and Long Short-term Memory for forecasting the demands, using large amounts of real historical data. In Phase 2, suppliers’ qualitative weights are determined by a fuzzy logic model. Then, a new multi-objective programming model is designed, considering multiple periods and products. In this phase, the results of Phase 1 and the results of the fuzzy model are utilized as inputs for the multi-objective model. The weighted-sum method is applied for solving the multi-objective model. The results show Random Forest model leads to more accurate predictions than the other examined models in this study. In addition, based on the results, the selection of the forecasting techniques and different weights of suppliers affect both supplier selection and the related orders.

{"title":"Machine learning techniques and multi-objective programming to select the best suppliers and determine the orders","authors":"Asma ul Husna, Saman Hassanzadeh Amin, Ahmad Ghasempoor","doi":"10.1016/j.mlwa.2025.100623","DOIUrl":"10.1016/j.mlwa.2025.100623","url":null,"abstract":"<div><div>Selection of appropriate suppliers and allocation the orders among them have become the two key strategic decisions regarding purchasing. In this study, a two-phase integrated approach is proposed for solving supplier selection and order allocation problems. Phase 1 contains four techniques from statistics and Machine Learning (ML), including Auto-Regressive Integrated Moving Average, Random Forest, Gradient Boosting Regression, and Long Short-term Memory for forecasting the demands, using large amounts of real historical data. In Phase 2, suppliers’ qualitative weights are determined by a fuzzy logic model. Then, a new multi-objective programming model is designed, considering multiple periods and products. In this phase, the results of Phase 1 and the results of the fuzzy model are utilized as inputs for the multi-objective model. The weighted-sum method is applied for solving the multi-objective model. The results show Random Forest model leads to more accurate predictions than the other examined models in this study. In addition, based on the results, the selection of the forecasting techniques and different weights of suppliers affect both supplier selection and the related orders.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"19 ","pages":"Article 100623"},"PeriodicalIF":0.0,"publicationDate":"2025-01-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143168591","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

S&P-500 vs. Nasdaq-100 price movement prediction with LSTM for different daily periods

Machine learning with applications

Pub Date : 2025-01-13 DOI: 10.1016/j.mlwa.2024.100617

Xiang Zhang, Eugene Pinsky

This paper explores the efficiency of LSTM neural networks in predicting price movements for the two major U.S. stock indices: the S&P-500 and the Nasdaq-100 index. We consider three distinct daily periods: “overnight” (Close-to-Open), “daytime” (Open-to-Close) and “24-hour” (Close-to-Close) trading sessions. Using historical pricing data for these indices since 2000, this study shows how well the standard LSTM model captures price movement patterns to improve short-term trading strategies. The findings reveal that, for the S&P-500, a one-year training with 24-hour periods delivers a 14.5% more return over the Buy-and-Hold strategy. Moreover, combining “overnight” and “daytime” strategies delivers more than 40% return compared to passive index investing. By contrast, for the Nasdaq-100, a shorter training period of three months for “24-hour” periods delivers 90% more return than passive index investing. These results suggest that LSTM effectively learns the unique market dynamics associated with each index and different time periods, offering further insights into how deep learning can enhance financial forecasting and trading opportunities.

{"title":"S&P-500 vs. Nasdaq-100 price movement prediction with LSTM for different daily periods","authors":"Xiang Zhang, Eugene Pinsky","doi":"10.1016/j.mlwa.2024.100617","DOIUrl":"10.1016/j.mlwa.2024.100617","url":null,"abstract":"<div><div>This paper explores the efficiency of LSTM neural networks in predicting price movements for the two major U.S. stock indices: the S&P-500 and the Nasdaq-100 index. We consider three distinct daily periods: “overnight” (Close-to-Open), “daytime” (Open-to-Close) and “24-hour” (Close-to-Close) trading sessions. Using historical pricing data for these indices since 2000, this study shows how well the standard LSTM model captures price movement patterns to improve short-term trading strategies. The findings reveal that, for the S&P-500, a one-year training with 24-hour periods delivers a 14.5% more return over the Buy-and-Hold strategy. Moreover, combining “overnight” and “daytime” strategies delivers more than 40% return compared to passive index investing. By contrast, for the Nasdaq-100, a shorter training period of three months for “24-hour” periods delivers 90% more return than passive index investing. These results suggest that LSTM effectively learns the unique market dynamics associated with each index and different time periods, offering further insights into how deep learning can enhance financial forecasting and trading opportunities.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"19 ","pages":"Article 100617"},"PeriodicalIF":0.0,"publicationDate":"2025-01-13","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143170230","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Multi-agent deep reinforcement learning with online and fair optimal dispatch of EV aggregators

Machine learning with applications

Pub Date : 2025-01-09 DOI: 10.1016/j.mlwa.2025.100620

Arian Shah Kamrani , Anoosh Dini , Hanane Dagdougui , Keyhan Sheshyekani

The growing popularity of electric vehicles (EVs) and the unpredictable behavior of EV owners have attracted attention to real-time coordination of EVs charging management. This paper presents a hierarchical structure for charging management of EVs by integrating fairness and efficiency concepts within the operations of the distribution system operator (DSO) while utilizing a multi-agent deep reinforcement learning (MADRL) framework to tackle the complexities of energy purchasing and distribution among EV aggregators (EVAs). At the upper level, DSO calculates the maximum allowable power for each EVA based on power flow constraints to ensure grid safety. Then, it finds the optimal efficiency-Jain tradeoff (EJT) point, where it sells the highest energy amount while ensuring equitable energy distribution. At the lower level, initially, each EVA acts as an agent employing a double deep Q-network (DDQN) with adaptive learning rates and prioritized experience replay to determine optimal energy purchases from the DSO. Then, the real-time smart dispatch (RSD) controller prioritizes EVs for energy dispatch based on relevant EVs information. Findings indicate the proposed enhanced DDQN outperforms deep deterministic policy gradient (DDPG) and proximal policy optimization (PPO) in cumulative rewards and convergence speed. Finally, the framework’s performance is evaluated against uncontrolled charging and the first come first serve (FCFS) scenario using the 118-bus distribution system, demonstrating superior performance in maintaining safe operation of the grid while reducing charging costs for EVAs. Additionally, the framework’s integration with renewable energy sources (RESs), such as photovoltaic (PV), demonstrates its potential to enhance grid reliability.

{"title":"Multi-agent deep reinforcement learning with online and fair optimal dispatch of EV aggregators","authors":"Arian Shah Kamrani , Anoosh Dini , Hanane Dagdougui , Keyhan Sheshyekani","doi":"10.1016/j.mlwa.2025.100620","DOIUrl":"10.1016/j.mlwa.2025.100620","url":null,"abstract":"<div><div>The growing popularity of electric vehicles (EVs) and the unpredictable behavior of EV owners have attracted attention to real-time coordination of EVs charging management. This paper presents a hierarchical structure for charging management of EVs by integrating fairness and efficiency concepts within the operations of the distribution system operator (DSO) while utilizing a multi-agent deep reinforcement learning (MADRL) framework to tackle the complexities of energy purchasing and distribution among EV aggregators (EVAs). At the upper level, DSO calculates the maximum allowable power for each EVA based on power flow constraints to ensure grid safety. Then, it finds the optimal efficiency-Jain tradeoff (EJT) point, where it sells the highest energy amount while ensuring equitable energy distribution. At the lower level, initially, each EVA acts as an agent employing a double deep Q-network (DDQN) with adaptive learning rates and prioritized experience replay to determine optimal energy purchases from the DSO. Then, the real-time smart dispatch (RSD) controller prioritizes EVs for energy dispatch based on relevant EVs information. Findings indicate the proposed enhanced DDQN outperforms deep deterministic policy gradient (DDPG) and proximal policy optimization (PPO) in cumulative rewards and convergence speed. Finally, the framework’s performance is evaluated against uncontrolled charging and the first come first serve (FCFS) scenario using the 118-bus distribution system, demonstrating superior performance in maintaining safe operation of the grid while reducing charging costs for EVAs. Additionally, the framework’s integration with renewable energy sources (RESs), such as photovoltaic (PV), demonstrates its potential to enhance grid reliability.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"19 ","pages":"Article 100620"},"PeriodicalIF":0.0,"publicationDate":"2025-01-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143170229","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Predicting customer subscription in bank telemarketing campaigns using ensemble learning models

Machine learning with applications

Pub Date : 2025-01-07 DOI: 10.1016/j.mlwa.2025.100618

Michael Peter , Hawa Mofi , Said Likoko , Julius Sabas , Ramadhani Mbura , Neema Mduma

This study investigates the use of ensemble learning models bagging, boosting, and stacking to enhance the accuracy and reliability of predicting customer subscriptions in bank telemarketing campaigns. Recognizing the challenges posed by class imbalance and complex customer behaviors, we employ multiple ensemble techniques to build a robust predictive framework. Our analysis demonstrates that stacking models achieve the best overall performance, with an accuracy of 91.88% and an Receiver Operating Characteristic Area Under the Curve (ROC-AUC) score of 0.9491, indicating a strong capability to differentiate between subscribers and non-subscribers. Additionally, feature importance analysis reveals that contact duration, economic indicators like the Euro interbank offered (Euribor) rate, and customer age are the most influential factors in predicting subscription likelihood. These findings suggest that by focusing on customer engagement and economic trends, banks can improve telemarketing campaign effectiveness. We recommend the integration of advanced balancing techniques and real-time prediction systems to further enhance model performance and adaptability. Future work could explore deep learning models and interpretability techniques to gain deeper insights into customer behavior patterns. Overall, this study highlights the potential of ensemble models in predictive modeling for telemarketing, providing a data-driven foundation for more targeted and efficient customer acquisition strategies.

{"title":"Predicting customer subscription in bank telemarketing campaigns using ensemble learning models","authors":"Michael Peter , Hawa Mofi , Said Likoko , Julius Sabas , Ramadhani Mbura , Neema Mduma","doi":"10.1016/j.mlwa.2025.100618","DOIUrl":"10.1016/j.mlwa.2025.100618","url":null,"abstract":"<div><div>This study investigates the use of ensemble learning models bagging, boosting, and stacking to enhance the accuracy and reliability of predicting customer subscriptions in bank telemarketing campaigns. Recognizing the challenges posed by class imbalance and complex customer behaviors, we employ multiple ensemble techniques to build a robust predictive framework. Our analysis demonstrates that stacking models achieve the best overall performance, with an accuracy of 91.88% and an Receiver Operating Characteristic Area Under the Curve (ROC-AUC) score of 0.9491, indicating a strong capability to differentiate between subscribers and non-subscribers. Additionally, feature importance analysis reveals that contact duration, economic indicators like the Euro interbank offered (Euribor) rate, and customer age are the most influential factors in predicting subscription likelihood. These findings suggest that by focusing on customer engagement and economic trends, banks can improve telemarketing campaign effectiveness. We recommend the integration of advanced balancing techniques and real-time prediction systems to further enhance model performance and adaptability. Future work could explore deep learning models and interpretability techniques to gain deeper insights into customer behavior patterns. Overall, this study highlights the potential of ensemble models in predictive modeling for telemarketing, providing a data-driven foundation for more targeted and efficient customer acquisition strategies.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"19 ","pages":"Article 100618"},"PeriodicalIF":0.0,"publicationDate":"2025-01-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143170241","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Improving mango ripeness grading accuracy: A comprehensive analysis of deep learning, traditional machine learning, and transfer learning techniques

Machine learning with applications

Pub Date : 2025-01-06 DOI: 10.1016/j.mlwa.2025.100619

Md․ Saon Sikder, Mohammad Shamsul Islam, Momenatul Islam, Md․ Suman Reza

Bangladesh ranks among the top 10 countries globally in mango output. Mangoes can be classified based on their ripeness, with skin color being the most significant aspect. The current classification procedure is done manually, leading to mistakes and vulnerability to human error. Most research often focuses on using a single method to assess the ripeness of fruits. The study comprises a set of comprehensive tests showcasing different tactics for determining the most efficient methods through various models. One unique dataset was used for all five models: Gaussian Naive Bayes (GNB), Support Vector Machine (SVM), Gradient Boosting (GB), Random Forest (RF), and K-Nearest Neighbors (KNN). Utilizing convolutional neural networks (CNNs) and VGG16, a pre-trained CNN model, to extract features and train the dataset. Used these training datasets as input to calculate the average accuracy of the five models during testing. In addition to these experiments, these five models using standard techniques also evaluated. The study also included a comparative analysis that emphasized the best performance of each model in various scenarios. This analysis shows that the CNN model consistently performs better than the transfer learning model (VGG16) and classical machine learning methods. Except for the KNN and Naive Bayes scenarios, the VGG16 model achieved much higher accuracy compared to typical machine learning methods. In three other models, classical machine learning outperforms the VGG16 model. The Gradient Boosting model in deep learning (CNN) demonstrated the highest accuracy of 96.28 % compared to other models and techniques.

{"title":"Improving mango ripeness grading accuracy: A comprehensive analysis of deep learning, traditional machine learning, and transfer learning techniques","authors":"Md․ Saon Sikder, Mohammad Shamsul Islam, Momenatul Islam, Md․ Suman Reza","doi":"10.1016/j.mlwa.2025.100619","DOIUrl":"10.1016/j.mlwa.2025.100619","url":null,"abstract":"<div><div>Bangladesh ranks among the top 10 countries globally in mango output. Mangoes can be classified based on their ripeness, with skin color being the most significant aspect. The current classification procedure is done manually, leading to mistakes and vulnerability to human error. Most research often focuses on using a single method to assess the ripeness of fruits. The study comprises a set of comprehensive tests showcasing different tactics for determining the most efficient methods through various models. One unique dataset was used for all five models: Gaussian Naive Bayes (GNB), Support Vector Machine (SVM), Gradient Boosting (GB), Random Forest (RF), and K-Nearest Neighbors (KNN). Utilizing convolutional neural networks (CNNs) and VGG16, a pre-trained CNN model, to extract features and train the dataset. Used these training datasets as input to calculate the average accuracy of the five models during testing. In addition to these experiments, these five models using standard techniques also evaluated. The study also included a comparative analysis that emphasized the best performance of each model in various scenarios. This analysis shows that the CNN model consistently performs better than the transfer learning model (VGG16) and classical machine learning methods. Except for the KNN and Naive Bayes scenarios, the VGG16 model achieved much higher accuracy compared to typical machine learning methods. In three other models, classical machine learning outperforms the VGG16 model. The Gradient Boosting model in deep learning (CNN) demonstrated the highest accuracy of 96.28 % compared to other models and techniques.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"19 ","pages":"Article 100619"},"PeriodicalIF":0.0,"publicationDate":"2025-01-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143170240","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Stairway to heaven: An emotional journey in Divina Commedia with threshold-based Naïve Bayes classifier

Machine learning with applications

Pub Date : 2024-12-24 DOI: 10.1016/j.mlwa.2024.100613

Maurizio Romano, Claudio Conversano

Computational literary uses data science and computer science techniques to study literature. In this framework, we investigate how an expert system can acquire knowledge from the specific content of a narrative text without any pre-existing information about it. We utilize the Threshold-based Naïve Bayes (Tb-NB) classifier to analyze the content of Dante Alighieri’s Divina Commedia poem. Tb-NB is a probabilistic data-driven model that predicts the polarity of a binary response based on the probability of an event occurring given certain features, and assigns a log-likelihood score to each word in a text. Our first task is understanding if and how the links between lexical forms and meanings characterize the three parts of the poem (Inferno, Purgatorio and Paradiso) in order to predict if a Canto belongs to Inferno or Paradiso based on its specific content, and to determine if a Canto of Purgatorio is more similar to those of Inferno or to those of Paradiso. We show Tb-NB outperform other similar approaches and achieves the same performance of Random Forest (F1-score

=

0.985) but providing much more information to interpret the specific content and the lexical forms used by Dante Alighieri in its poem. The Tb-NB’s scores are the base of knowledge for the implementation of an expert system, like a search engine, that can help users to identify the most informative verses of a Canto or by better comprehend or discover the content of the poem from a word related to a particular feeling or emotion.

{"title":"Stairway to heaven: An emotional journey in Divina Commedia with threshold-based Naïve Bayes classifier","authors":"Maurizio Romano, Claudio Conversano","doi":"10.1016/j.mlwa.2024.100613","DOIUrl":"10.1016/j.mlwa.2024.100613","url":null,"abstract":"<div><div>Computational literary uses data science and computer science techniques to study literature. In this framework, we investigate how an expert system can acquire knowledge from the specific content of a narrative text without any pre-existing information about it. We utilize the Threshold-based Naïve Bayes (Tb-NB) classifier to analyze the content of Dante Alighieri’s Divina Commedia poem. Tb-NB is a probabilistic data-driven model that predicts the polarity of a binary response based on the probability of an event occurring given certain features, and assigns a log-likelihood score to each word in a text. Our first task is understanding if and how the links between lexical forms and meanings characterize the three parts of the poem (Inferno, Purgatorio and Paradiso) in order to predict if a Canto belongs to Inferno or Paradiso based on its specific content, and to determine if a Canto of Purgatorio is more similar to those of Inferno or to those of Paradiso. We show Tb-NB outperform other similar approaches and achieves the same performance of Random Forest (F1-score <span><math><mo>=</mo></math></span> 0.985) but providing much more information to interpret the specific content and the lexical forms used by Dante Alighieri in its poem. The Tb-NB’s scores are the base of knowledge for the implementation of an expert system, like a search engine, that can help users to identify the most informative verses of a Canto or by better comprehend or discover the content of the poem from a word related to a particular feeling or emotion.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"19 ","pages":"Article 100613"},"PeriodicalIF":0.0,"publicationDate":"2024-12-24","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143170231","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0

Unified wound diagnostic framework for wound segmentation and classification

Machine learning with applications

Pub Date : 2024-12-20 DOI: 10.1016/j.mlwa.2024.100616

Mustafa Alhababi , Gregory Auner , Hafiz Malik , Muteb Aljasem , Zaid Aldoulah

Chronic wounds affect millions worldwide, posing significant challenges for healthcare systems and a heavy economic burden globally. The segmentation and classification (S&C) of chronic wounds are critical for wound care management and diagnosis, aiding clinicians in selecting appropriate treatments. Existing approaches have utilized either traditional machine learning or deep learning methods for S&C. However, most focus on binary classification, with few addressing multi-class classification, often showing degraded performance for pressure and diabetic wounds. Wound segmentation has been largely limited to foot ulcer images, and there is no unified diagnostic tool for both S&C tasks. To address these gaps, we developed a unified approach that performs S&C simultaneously. For segmentation, we proposed Attention-Dense-UNet (Att-d-UNet), and for classification, we introduced a feature concatenation-based method. Our framework segments wound images using Att-d-UNet, followed by classification into one of the wound types using our proposed method. We evaluated our models on publicly available wound classification datasets (AZH and Medetec) and segmentation datasets (FUSeg and AZH). To test our unified approach, we extended wound classification datasets by generating segmentation masks for Medetec and AZH images. The proposed unified approach achieved 90% accuracy and an 86.55% dice score on the Medetec dataset and 81% accuracy and an 86.53% dice score on the AZH dataset These results demonstrate the effectiveness of our separate models and unified approach for wound S&C.

{"title":"Unified wound diagnostic framework for wound segmentation and classification","authors":"Mustafa Alhababi , Gregory Auner , Hafiz Malik , Muteb Aljasem , Zaid Aldoulah","doi":"10.1016/j.mlwa.2024.100616","DOIUrl":"10.1016/j.mlwa.2024.100616","url":null,"abstract":"<div><div>Chronic wounds affect millions worldwide, posing significant challenges for healthcare systems and a heavy economic burden globally. The segmentation and classification (S&C) of chronic wounds are critical for wound care management and diagnosis, aiding clinicians in selecting appropriate treatments. Existing approaches have utilized either traditional machine learning or deep learning methods for S&C. However, most focus on binary classification, with few addressing multi-class classification, often showing degraded performance for pressure and diabetic wounds. Wound segmentation has been largely limited to foot ulcer images, and there is no unified diagnostic tool for both S&C tasks. To address these gaps, we developed a unified approach that performs S&C simultaneously. For segmentation, we proposed Attention-Dense-UNet (Att-<span>d</span>-UNet), and for classification, we introduced a feature concatenation-based method. Our framework segments wound images using Att-<span>d</span>-UNet, followed by classification into one of the wound types using our proposed method. We evaluated our models on publicly available wound classification datasets (AZH and Medetec) and segmentation datasets (FUSeg and AZH). To test our unified approach, we extended wound classification datasets by generating segmentation masks for Medetec and AZH images. The proposed unified approach achieved 90% accuracy and an 86.55% dice score on the Medetec dataset and 81% accuracy and an 86.53% dice score on the AZH dataset These results demonstrate the effectiveness of our separate models and unified approach for wound S&C.</div></div>","PeriodicalId":74093,"journal":{"name":"Machine learning with applications","volume":"19 ","pages":"Article 100616"},"PeriodicalIF":0.0,"publicationDate":"2024-12-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143170239","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

引用次数: 0