Comparing Discriminant Analysis Function for Early Prediction of Smartphone Addiction

The pervasive use of smartphones in daily life has led to significant benefits, but excessive use has caused alarming behavioral and health issues, particularly among adolescents. Addressing smartphone addiction requires early detection to enable timely interventions. This study investigates the application of machine learning, specifically Linear Discriminant Analysis (LDA) and Quadratic Discriminant Analysis (QDA), for the early prediction of smartphone addiction. The research used a dataset containing 394 instances categorized into "addicted" and "non-addicted" classes. Dataset is derived from questionnaire responses. After preprocessing steps, including feature selection and ordinal encoding, the data was split using 10-fold cross-validation to ensure robust evaluation. The models were assessed using metrics such as accuracy, precision, recall, and F-measure. Results indicate that LDA significantly outperforms QDA across all metrics, achieving an accuracy of 94.16%, a precision of 94.2%, a recall of 94.2%, and an F-measure of 94.2%. Additionally, the Receiver Operating Characteristic (ROC) curve analysis showed an Area Under the Curve (AUC) of 0.9875 for LDA, indicating its high reliability and stability in classifying smartphone addiction. QDA, while effective, has a slightly lower performance due to the linear separability of the dataset. This study concludes that LDA is a robust and effective method for early prediction of smartphone addiction, offering valuable insights for health monitoring systems. The findings provide a foundation for future applications of discriminant analysis in addressing behavioral health issues.