2.1 Using Artificial Intelligence to Identify Risk of School Violence

Abstract

A standardized risk assessment protocol was established in order to evaluate risk of violence by children in the school setting. This was tested through the use of natural language processing (NLP) and machine learning (ML) technologies in order to evaluate the accuracy of an artificial intelligence (AI) previously created by our team. The aim of this project is to develop an AI that can be used in schools to screen and help prevent violence. The Brief Rating of Aggression by Children and Adolescents (BRACHA) and the School Safety Scale (SSS) were used to interview and collect data on 396 participants who were between the ages of 10 and 18 years and enrolled in school. Risk of violence was determined by a violence assessment team based on clinical judgment, and separately by AI. The AI used NLP technologies to extract different types of linguistic features from the interview. ML classifiers were then applied to predict risk of school violence for individual subjects. Using statistical methods, clinical judgement was compared to the AI. Clinical assessments of moderate and high risk were grouped and are both considered a strong match to the algorithm’s high-risk category (area under the curve [AUC] = 0.9500). Using the random forest algorithm, the BRACHA was determined to be the best predictor of aggression, followed by the SSS and then by demographic data (age at time of visit, followed by number of adults in the household, and then by number of children in the household [p < 0.01]). The high AUC shows that our AI is effective in assessing the risk of violence in the school setting. Linguistic features significantly outperformed subject household information in predicting violence. NLP and ML classifiers show strong promise in detecting students at risk of perpetrating school violence. With a greater study population and further developed computerized algorithms, we are confident that an AI for preventing any and all school violence can be created.