Predicting problematic smartphone use based on early maladaptive schemas by using machine learning classification algorithms

This study aimed to predict problematic smartphone use based on early maladaptive schemas (EMS) and five schema domains. Machine learning algorithms were used to test the predictive models based on data collected from 1000 smartphone users. The study tested the predictive models by employing six machine learning classification algorithms (i.e., Bayes Net, SMO, IBk, Multi-Class Classifier, Decision Table, and Random Forest). The first predictive model was built on 14 schemas and tested by using 10-fold cross validation method. Results indicated that Multi-Class Classifier achieve a better prediction than other classifiers in classifying low-risk and high-risk smartphone users based on 14 schemas with an accuracy of 68.2%. The second model, which was built on five schema domains, was tested by using the best performance algorithm. Multi-Class Classifier predicted the users based on schemas related to the "impaired autonomy and performance" domain (i.e., "enmeshment/dependence, vulnerability to harm, and failure") with an accuracy of 66.2%. Further, the classifier predicted the users based on schemas related to the "disconnection and rejection" domain (i.e., "abandonment, emotional deprivation, defectiveness, and social isolation/mistrust") with an accuracy of 65.2%. Results emphasize the significance of EMS in predicting problematic smartphone use.