Predictive Analytics Project

Crime Finder

A Predictive Analytics Machine Learning Project.
Analyzes national crime data and develops models to predict the likelihood that a specific criminal offense will occur using k-NN and Random Forest algorithms.

Executive Summary

This study sheds light on current events by analyzing crime statistics. Historical data was analyzed and a model built to predict situations and locations in which specific offenses are likely to occur, such as justifiable homicide. The business objective of this study is to lower the number of justifiable homicides. Predictions will show the government, community, and media where changes need to be made the most. Funding can be raised in at-risk areas, increasing the sense of community and providing an added sense of protection to all. The source and application of funding is outside the scope of this project. The model objective was to create a model in which the predicted output is the likelihood that a specific criminal offense will occur. This document describes the technical methods used in building the predictive model and model evaluation results. The final model predicts the type of offense that will occur with an 84% accuracy rate, given geographic and victim demographic information. This model can be used for predictions for any location across the United States. A threshold can be set to redirect focus for funding and training. For example, if a city is predicted to have over two (2) justifiable homicides per 10,000 people, you can take action for change.

Introduction

Protests throughout the country have shed light on racial profiling, excessive use of force, and unjust biases. The publicly broadcast death of George Floyd brought forth questions around the world about police tactics and whether they are justified. Throughout the years, time and again, unjust incidents have brought up the question of systemic racism.

Business Problem

Too many reports of Excessive Force. There has been an increase in reported incidents of excessive use of force by police. This increase has heightened visibility, and communities across the country are calling for change. The police have a reputation for targeting minorities, making some afraid to call the police when needed. Change is needed to deter crime while providing a sense of protection to the entire community.

Data Sources

Uniform Crime Reporting Program Data: National Incident-Based Reporting System (NBIRS), 2016; United States Federal Bureau of Investigation; Inter-university Consortium for Political and Social Research (ICPSR), University of Michigan; https://www.icpsr.umich.edu/icpsrweb/NACJD/NIBRS/

US Cities Database: https://simplemaps.com/data/us-cities State Geodetic Centers: https://www.latlong.net

Modeling Objective

The predicted output of the model should be the likelihood that a specific criminal offense will occur.

Click here to read on for Methods and Results
Click here if you’d rather see a presentation
Click here to access Source Code