The Aberrant Behavioral Spectrum of Internet Addiction in University Students

A large number of behaviors, collected by self-reporting questionnaires, have been demonstrated associative with Internet addiction (IA). Assume that these behaviors are informative for distinguishing IA subjects from normal controls (NCs). Rather than analyzing each behavior separately, a classification model is presented for discriminant analysis of behaviors jointly at a subject level. In conjunction with support vector machine (SVM) classifier, a forward behavior selection technique is presented to select behaviors for constructing the most discriminative behavioral spectrum. The discriminant analysis model has been applied to 8 types of behaviors based on IA study with 27 IA subjects and 27 NCs. These 8 types of behaviors were average online days per week, average online time per day, anxiety level, depression level, self-control ability, positive self-confidence, negative self-confidence, and self-confidence, which were closely associated with AI. Experimental results indicated that the aberrant behavioral spectrum were properly arranged, which identified behavioral markers with different levels of importance for IA. The presented classification model also achieved an excellent performance for differentiating IA subjects from NCs (average area under the receiver operating characteristic curve [AUROC]: 0.96, average accuracy: 84%, average specificity: 92%, average sensitivity: 76%).