This study was conducted to explore major predictors affecting real-life delinquency and cyber delinquency of middle school students. To this end, data from the fourth wave of the elementary school panel (first year of middle school) of the Korean Children and Youth Panel Survey 2018 (KCYPS 2018) conducted by the Korea Youth Policy Institute was used. Random forest analysis was applied to derive major predictors and the relationship between the predictors with real-life delinquency and/or cyber delinquency. First, leisure time in playing with computers, academic helplessness-lack of active performance, attention, aggression, depression, and smartphone dependence were derived as main predictors for both real-life delinquency and cyber delinquency. Second, academic enthusiasm, physical symptoms, and positive friendship were negatively related to real-life delinquency, while leisure time in playing with friends was positively related. Third, learning time for Internet and TV lectures, learning time at private institutes and tutoring, and time spent with parents during weekdays were confirmed to have a negative relationship with cyber delinquency. Based on these research results, implications and suggestions for future research are presented to reduce and prevent real-life and cyber delinquency in middle school students.