A comprehensive analysis of the determinants influencing sexual assault crime rates across the United States using machine learning and explainable AI

Abstract Sexual assault (SA) crimes are horrific crimes against humanity. It is critical for our society to find ways to reduce and ultimately eradicate this issue. How-ever, it is not yet clear which of the possible factors impact the occurrence of SA crimes more than others. There is also a lack of quantitative studies on SA crime rates. This leads to less focused and targeted efforts when trying to reduce the occurrence of SA. This paper fills these gaps by leveraging machine learning techniques. The work explores the impact of various legal, societal, economic, and demographic factors across all 50 US states on SA crime. A Random For-est regression model is used to determine the relative influence of each factor on SA crime rates. SHAP values are used to identify the features that correlate the most and least with SA crime rates. The work suggests that factors such as drug usage and rural population density are more strongly correlated with SA crime rates than legal factors such as the strictness of statute penalties. The findings indicate that rather than focusing solely on increasing statutory strictness, directing efforts to other solutions, such as educational programs and community initiatives, will be more impactful. The study also contradicts conclusions from previous studies by finding that states with more registered sex offenders have lower SA crime rates.