Analyzing Digital Addiction through Feature Selection and Machine Learning Techniques
Saved in:
| Title: | Analyzing Digital Addiction through Feature Selection and Machine Learning Techniques |
|---|---|
| Language: | English |
| Authors: | Mizanur Rahman, Hasan Sarwar, Md Kamrul Hasan, Ting Tin Tin |
| Source: | Educational Process: International Journal. Article e2026067 2026 23. |
| Availability: | UNIVERSITEPARK Limited. iTOWER Plaza (No61, 9th floor) Merkez Mh Akar Cd No3, Sisli, Istanbul, Turkey 34382. e-mail: editor@edupij.com; Web site: http://www.edupij.com/ |
| Peer Reviewed: | Y |
| Page Count: | 24 |
| Publication Date: | 2026 |
| Document Type: | Journal Articles Reports - Research |
| Descriptors: | Addictive Behavior, Prediction, Artificial Intelligence, Statistical Analysis, Social Media, Students |
| ISSN: | 2147-0901 2564-8020 |
| Abstract: | Background/purpose: Digital addiction has become an increasing concern, particularly among students, as excessive engagement with digital platforms negatively impacts academic performance, social interactions, and mental health. This study primarily aims to evaluate and compare the effectiveness of various machine learning and deep learning models in predicting digital addiction. While we examine key features that contribute to classification performance, our focus is not on discovering new predictors but on selecting features that enhance model accuracy. However, few datasets are publicly available, and, moreover, these datasets do not employ similar feature types. As a result, it is not readily clear which set of features a researcher should consider to obtain the most accurate prediction. Materials/methods: We utilized two publicly available datasets and conducted feature selection using ANOVA scores and Random Forest. We then performed prediction of digital addiction using four machine learning (ML) models--K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting (GB)--as well as one deep learning model, Artificial Neural Network (ANN). These models were evaluated using standard performance metrics, including accuracy, precision, recall, and F1 Score. Results: Our results show that Random Forest, when used to identify the most important feature, performs better overall, especially for models such as KNN. The Random Forest and Gradient Boosting models achieved perfect scores (100% accuracy, precision, recall, and F1-score) on one dataset, and the ANN model reached up to 99.5% accuracy. Conclusion: Random Forest-based feature importance is particularly effective, and the overall results suggest that ensemble models such as RF and GB are highly reliable. This research will guide future researchers, as they create their own dataset, on the importance of selecting appropriate features in combination with machine learning algorithms to achieve better predictive accuracy. These insights may support early interventions in educational settings to reduce the academic risks associated with digital addiction. |
| Abstractor: | As Provided |
| Entry Date: | 20260101 |
| Accession Number: | EJ1514572 |
| Database: | ERIC |
| Abstract: | Background/purpose: Digital addiction has become an increasing concern, particularly among students, as excessive engagement with digital platforms negatively impacts academic performance, social interactions, and mental health. This study primarily aims to evaluate and compare the effectiveness of various machine learning and deep learning models in predicting digital addiction. While we examine key features that contribute to classification performance, our focus is not on discovering new predictors but on selecting features that enhance model accuracy. However, few datasets are publicly available, and, moreover, these datasets do not employ similar feature types. As a result, it is not readily clear which set of features a researcher should consider to obtain the most accurate prediction. Materials/methods: We utilized two publicly available datasets and conducted feature selection using ANOVA scores and Random Forest. We then performed prediction of digital addiction using four machine learning (ML) models--K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Random Forest (RF), and Gradient Boosting (GB)--as well as one deep learning model, Artificial Neural Network (ANN). These models were evaluated using standard performance metrics, including accuracy, precision, recall, and F1 Score. Results: Our results show that Random Forest, when used to identify the most important feature, performs better overall, especially for models such as KNN. The Random Forest and Gradient Boosting models achieved perfect scores (100% accuracy, precision, recall, and F1-score) on one dataset, and the ANN model reached up to 99.5% accuracy. Conclusion: Random Forest-based feature importance is particularly effective, and the overall results suggest that ensemble models such as RF and GB are highly reliable. This research will guide future researchers, as they create their own dataset, on the importance of selecting appropriate features in combination with machine learning algorithms to achieve better predictive accuracy. These insights may support early interventions in educational settings to reduce the academic risks associated with digital addiction. |
|---|---|
| ISSN: | 2147-0901 2564-8020 |
