Machine Learning, Supervised Leveraging Supervised Machine Learning to Improve Accuracy in Predicting Smartphone Addiction

Abstract

The transformation of smartphones into essential life companions triggers new challenges in the form of digital addiction that erodes productivity and cognitive focus. In contrast to previous studies that focused on screen duration, this study explored a more critical dimension, namely the degradation of productive hours. The research aims to improve the prediction of addiction using supervised machine learning by integrating productivity variables as a core diagnostic feature. Through a database of 7,500 respondents, a comparative analysis was carried out on  the Random Forest, K-Nearest Neighbors (KNN), and Naïve Bayes algorithms in The results of the study proved that the productivity variable significantly increased the predictability of the model. Random Forest consistently outperformed other algorithms across scenarios, with the highest accuracy reaching 93.32% at a 70:30 data ratio, surpassing KNN (89.66%) and Naïve Bayes (87.96%). With a specificity of 92.70%, these findings reveal that workflow disruption has a higher differentiating power than conventional duration metrics. In conclusion, productivity interference is a key indicator in identifying digital behavioral disorders. In practical terms, this research provides guidance for developers and managers in designing precise early warning systems to restore daily efficiency amid the distractions of the digital world.