Análise de incidentes de segurança usando séries temporais e modelos ARIMA

Abstract

This work uses data analysis concepts from an independently maintained public database on security incidents, the Hackmageddon, in order to try to predict future behavior of the same. During the research, the data were transformed into five time series, one for all incidents and four for each type of incident (Cyber Crime, Cyber Espionage, Cyber War and Hacktivism). From them, autocorrelation functions, partial autocorrelation functions, graph analysis and validation tests were used to build ARIMA models and elaborate predictions from them. The results indicated a similar performance among the models that had longer estimation periods performed better than those thatused smaller periods.