Interpretable Prediction Method of Adolescent Internet Addiction Based on DBO-XGBoost

Adolescent Internet addiction is a global social problem in the Internet age, which seriously affects the physical and mental health of adolescents. Exploring its causes and accurately predicting addiction individuals is a current research hotspot. At present, most of the researches on predicting Internet addiction based on machine learning only consider a small number of addiction characteristics, and there are shortcomings in the accuracy and interpretability of the prediction results. Therefore, this paper proposes an interpretable prediction method for Internet addiction. Firstly, based on the psychological I-PACE framework, a multi-factor dataset of Internet addiction is constructed. Then, the XGBoost model is used to predict Internet addiction, and the parameters of the model are optimized by DBO to improve the prediction accuracy of the model. Finally, SHAP model is used to explain the main causes of Internet addiction from two aspects of the global and individual, which improves the interpretability of the model. Experimental results show that the accuracy of DBO-XGBoost model prediction reaches 87.82%, and the AUC reaches 93.13%, and the interpretability of prediction results is enhanced by using SHAP method.