Abstract: Predictors of Offline and Online Delinquency: A Machine Learning Approach (Society for Social Work and Research 27th Annual Conference - Social Work Science and Complex Problems: Battling Inequities + Building Solutions)

All in-person and virtual presentations are in Mountain Standard Time Zone (MST).

SSWR 2023 Poster Gallery: as a registered in-person and virtual attendee, you have access to the virtual Poster Gallery which includes only the posters that elected to present virtually. The rest of the posters are presented in-person in the Poster/Exhibit Hall located in Phoenix A/B, 3rd floor. The access to the Poster Gallery will be available via the virtual conference platform the week of January 9. You will receive an email with instructions how to access the virtual conference platform.

572P Predictors of Offline and Online Delinquency: A Machine Learning Approach

Sunday, January 15, 2023
Phoenix C, 3rd Level (Sheraton Phoenix Downtown)
* noted as presenting author
Taekho Lee, MA, Doctoral student, Seoul National University, Seoul, Korea, Republic of (South)
Yeonjeong Seo, BA, Student-Master's, Seoul National University, Seoul, Korea, Republic of (South)
Jisu Park, PhD, Postdoctoral Researcher, Seoul National University, Seoul, Korea, Republic of (South)
Yoonsun Han, PhD, Associated Professor, Seoul National University, Seoul, Korea, Republic of (South)

With the recent surge in adolescent online activity, the problem of online delinquency has surfaced as an important social issue. In this changing social context, researchers and practitioners have raised a fundamental issue: Whether major predictors that explain traditional juvenile delinquency are also applicable for understanding online delinquency. Identifying the similarities and differences between traditional offline and emergent online delinquency may provide directions for intervention and development of coping strategies for adolescents. The goal of this study is to verify and compare the predictors of offline and online delinquency using a random forest machine learning approach. Specifically, this study asks: What are the predictors of offline and online juvenile delinquency? What are the similarities and differences between these predictors?


This study used nationally representative data from the International Self Report Delinquency Study (2021-2022) to investigate online and offline delinquency experiences of Korean adolescents (N=1,355). A random forest model was examined to identify predictor variables that play an important role in predicting online and offline delinquency, respectively. A total of 79 predictor variables comprising individual, family, peer, and school domains that are theoretically relevant to delinquency were included in the model. The data were separated into training data and validation data at a ratio of 7:3, and 5-fold cross-validation was repeated 3 times in the process of building a model with the training data. The ROC-AUC was used as the evaluation index of the model. Afterwards, the relationship between the outcome variable and the predictor variable was depicted using the partial dependence plot for the top 5 important variables.


The predictive performance of the online and offline delinquency models were reliable (AUC=0.84; AUC=0.80). Main variables as suggested by random forest results were examined. First, the variables that predict offline delinquency are, in order of importance, ‘school environment’, ‘impulsivity’, ‘number of friends’, ‘revenge’, ‘prosocial ability’. Next, for online delinquency, ‘cyber violence victimization’, ‘negative parent-child relationship’, ‘online privacy violations committed by friends’, ‘violence perception’, ‘revenge’ were shown in order. Greater impulsivity, number of friends, purpose of revenge, cyber violence victimization, negative parent-child relationship, and online privacy violations committed by friends had a positive marginal effect, whereas a positive school environment and negative perception toward violence had a negative marginal effect on delinquency.

Conclusions and Implications:

The current study identified predictors of offline and online delinquency and compared similarities and differences. Offline delinquency was mostly influenced by ecological factors such as school and friends, whereas online delinquency was affected by behavioral or cognitive factors such as victimization experience and perception of violence. In other words, in predicting online delinquency, adolescents’ personal experience seem to be more important than the direct influence of surrounding social environments. Such results may be related with unique characteristics of online environments such as disembodiment or non-face-to-face. These findings suggest the need for applying unique strategies that are distinct from traditional ones when addressing the issue of online delinquency in the new era of information and technology.