Studying the factors that influence the Incidence of UCR Part One Crimes in Boston

Abstract

Boston, as one of the most populated cities in the United States, while being known for its rich history, humanities and culture, also faces significant challenges with crime. This paper presents research on the critical factors that influence a possibility of a crime in Boston being UCR part 1 crime. If applied to law enforcement agencies, the distribution of police resources in Boston can be more reasonable and enable better urban security. We obtained the dataset from a Kaggle post, which was obtained from the Crime Incident Report of the Boston Police Department. By cleaning the data, converting months to seasons, we made the data more usable. We then split the data in a 70-30% ratio. We ran two models – first for the training session, we used the random forest classifier to identify the most significant factors. We evaluated the model by using metrics such as accuracy, precision, recall, and the ROC-AUC curve. In the testing session, we employed a logistic regression model for cross validation. For both models, the obtained accuracy is over 94%, indicating an extremely high overall performance for both models. Through the data analysis and test in this paper, it can be summarized that both the logistic regression model and random forest classifier can effectively predict and analyze an instance of a UCR part 1 crime in Boston based on the location of the crime, crime code group, season, time of the day, and day of the week. Though the accuracy is significantly high, we should not ignore the fact that we used a slightly older dataset ranging from 2015 to 2018, which, being six years ago, may have intrinsically different crime patterns than those that occur now.