Forecasting school violence risk with incomplete interview data: an automated assessment approach

Area of Science:

• Psychology
• Computer Science
• Criminology

Background:

• School violence risk assessment traditionally relies on manual, time-consuming, and subjective methods.
• Developing objective and efficient tools is crucial for effective prevention strategies.

Purpose of the Study:

• To develop and evaluate a machine learning algorithm, Automated RIsk Assessment (ARIA), for predicting aggression risk using NLP.
• To assess the algorithm's performance with incomplete interview data.

Main Methods:

• ARIA utilized natural language processing (NLP) on standardized interview questions to identify linguistic patterns predictive of aggression.
• Feature sets were incrementally added, simulating partial interviews, and evaluated using L2-regularized logistic regression and L2-SVM classifiers.
• Data included 412 interviews from the Brief Rating of Aggression by Children and Adolescents (BRACHA) and School Safety Scale (SSS) instruments.

Main Results:

• ARIA achieved an area under the ROC curve of 0.9 after only 10 BRACHA questions, demonstrating high predictive power even with truncated interviews.
• The algorithm's performance remained robust, comparable to full assessments, when using incomplete data.
• The full BRACHA and BRACHA + SSS assessments showed similar performance levels.

Conclusions:

• ARIA offers a viable solution for risk assessment when interviews cannot be fully completed.
• The algorithm can reduce the burden on school personnel by providing reliable risk predictions from partial data.
• This technology has the potential to enhance school safety and violence prevention efforts.