Behaviour-Aware Explainable Artificial Intelligence Framework for Criminal Offender Profiling

1. Dr.K.Priya
Assistant Professor,
Department of Artificial Intelligence and Machine Learning,
SoISDS, KPRCAS, Coimbatore
2. Dr.K.Dheenathayalan
Assistant Professor, Department of Computer Science
NIFT-TEA College of Knitwear Fashion, Tirupur

Abstract: In recent years the use of artificial intelligence for criminal offender profiling has grown which at the same time has seen the2 issue of transparency in most predictive models which has not gone away still we have issues with belief in these models, with who is responsible when something goes wrong, and also the ethical play out of these in the criminal justice field. We don't see it as sufficient that a model is accurate in criminal justice settings; what we also requireis that the model’s decisions are also made clear to human stakeholders. That is what this paper sets out to do which is present an Explainable AI based framework in which we look at the behavioural patterns of the offender at risk instead of the dense stats. Also, we put forth a set of behaviour-based features which we get from past offense reports. Recidivism risk is estimated using ensemble-based machine learning models, and explainability techniques are then used to interpretthe predictions. These explanations offer both localised explanations for specific offender predictions and global perceptions of important behavioural factors. According to the experimental results, the suggested framework significantly improves model interpretability while achieving competitive predictive performance.
Keywords: Explainable Artificial Intelligence, Criminal Offender Profiling, Recidivism Prediction, BehaviouralAnalytics, SHAP and LIME.