Real-Time Detection of Internet Addiction Using Reinforcement Learning System

Abstract

Since Internet addiction (IA) was reported in 1996, research on IA assessment has attracted considerable interest. The development of a real-time detector system can help communities, educational institutes, or clinics immediately assess the risk of IA in Internet users. However, current questionnaires were designed to ask Internet users to self-report their Internet experiences for at least 6 months. Physiological measurements were used to assist questionnaires in the shortterm assessment of IA, but physiological properties cannot assess IA in real-time due to a lack of algorithms. Therefore, the real-time detection of IA is still a work in progress. In this study, we adopted an extended classifier system with continuous real-coded variables (XCSR), which can solve the non-Markovian problem with continuous real-values to produce optimal policy, and determine high-risk and low-risk IA using Chen Internet addiction scale (CIAS) data or respiratory instantaneous frequency (IF) components of Internet users as input information. The result shows that the classification accuracy of XCSR can reach close to 100%. We also used XCSR to verify the items of CIAS and extract important respiratory indexes to assess IA. We expect that a real-time detector that immediately assesses the risk of IA may be designed in this way.