ABSTRACT
This study researches cybercafé users’ behaviours with respect to Internet addiction diagnostics. What we do includes two steps. The first is survey design, and the second is getting knowledge from the swarm of questionnaire results. For the second step, representative variables are selected using variable clustering together with correlation analysis. To compensate the defect of class imbalance, we design a data resampling mechanism to obtain a balanced dataset. The prediction algorithm C4.5 decision tree is utilized to extract predictive variables from the large number of variables, and to generate explicit principles for Internet addiction prediction.
