Machine Learning Analysis of Gun Violence

Abstract

In the U.S, gun violence is a substantial and divisive issue that legislators have spent decades trying to solve by passing various laws, restrictions, and bans using data analysis and research to support their ideas. This paper aims to use machine learning models such as logistic regression, random forest classifiers, gradient boosting classifiers, (Support Vector Machine) SVC classifiers, and Multi-layer Perceptron (MLP) classifiers to analyze different pieces of shooting incidents. Using the feature weights of the best-performing model, we aim to find which factors have the largest effect on shooting incident lethality - whether anybody died or not during the incident. Based on this, lawmakers can focus on the factors that most contribute to lethality. The first models in the paper aim to predict whether a shooting was lethal or not based on the age and gender of the shooter(s), state poverty rate, percentage of people in that state with only a high school education rate, shooting location, year, etc. The other models aim to predict the shooter’s gender, the state that the shooting was in, and the political trifecta of that state based on similar factors. The precision, recall, and f-score of these models, as well as the feature weights of the random forest classifiers and gradient boosting classifiers, are reported. We used tree-based models to analyze the feature weights because of their high performance and because of Sk-learn’s limitations in finding feature weights for other model types. Additionally, we include an analysis and possible reasoning for our results.