Luster Regained : A Novel Cyber Incident Risk Prediction Model Using Machine Learning

Abstract

Physical isolation during the COVID-19 pandemic prompted a 45% increase in digital use [5,15], leading to an increase in cyber incidents. This project seeks to understand the risk impact of prolonged internet use and evaluate opportunities for cyber education to lower such risk. In preparation for subsequent work, the project will learn about patterns in distress and the recovery of affected individuals. A 20-question English-language survey (Appendix A) was completed by 6th through 12th graders (n=1,869) across 4 countries. Analysis of the survey [1, 8, 10, 11, 13] indicated that the number of hours of internet use was a driver of the risk of cyber incidents. In addition to statistical analysis, the methodology used Google’s VertexAI AutoML [6] to generate an ensemble model to predict risk (on n=1 basis) from usage patterns (length of usage, gaming use, etc.). The cyber risk predictor model set has high overall accuracy (f1 score of 0.88) and precision and recall of 0.878. This low-cost approach to personalized risk scores could support periodic evaluation and trending of educational effectiveness in cyber safety. Separately, participants reported a strong association (Spearman’s Rho = 0.957) between distress from cyber incidents and recovery time. Among the respondents with high distress experiences, there is an urgent need to design support programs to help them cope.