A multifaceted risk assessment approach using statistical learning to evaluate socio-environmental factors associated with regional felony and misdemeanor rates

Delinquencies are burden on a society, and minimizing the crime risk is essential. In this paper, we propose a multifaceted data-driven approach to long-term predictions of annual felony and misdemeanor rates and understanding their spatial patterns under various scenarios. Leveraging a suite of nonlinear statistical learning algorithms, we developed advanced models to predict county-level annual crime rates, and evaluate the associations of felony and misdemeanor rates with socioeconomic and demographic variables, infrastructure patterns in neighborhoods, and climatic variables. We implemented our proposed framework for the state of New York, the fourth most populous state in the U.S. For both felony and misdemeanor predictions, our results indicate that Random Forest outperforms all the other models achieving over 84% and 60% improvements in goodness-of-fit and predictive accuracy respectively, compared to the mean-only model. We identified population demography, socioeconomic condition, and infrastructure patterns to be key predictors of both felony and misdemeanor rates, with suburban crime rates being significantly higher than the urban and rural ones. Specifically, we observed that higher population count, higher poverty rate, and lower median family income are associated with elevated crime rates. Higher number of transport infrastructures, shopping malls and banks are also found to be positively correlated with increasing crime rates. In addition, we show how scenario-based sensitivity analysis can be leveraged to communicate crime risk to the stakeholders under various scenarios. Our proposed framework can help both policymakers and law enforcement in informed decision-making towards crime management, thereby minimizing risk of delinquencies in society at large.