CUSP-GX 5006.001 Course Project
Predicting the Felony Amount Based on Street Conditions and Demographics at Census Tract Level
Abstract
Researches on crimes can be "person-based" and "location-based". In this study, we explore the relationship between demographic (person-based), street condition(location-based) and felonies' amount. We collected data for the 288 census tracts on Manhattan, and used a set of data mining classification techniques. After analyzed the result to determine which is the best for explain how the demographic and street condition affect crime, we propose a best approach which classifies the crime hierarchy of 288 census tracts correctly more than 60% based on built environment and demographics.
Introduction
Understanding the causes of crime is important for a city’s safety control. In additional to demographic causes in that area, the perceived view of streetscapes is also the possible causes, due to the strong relationship between visual appearance and people’s behavior. While it’s hard to extract specific causality relationship, using machine learning algorithms to explore the relationship between urban data and crime amount has been proved as an effective approach. This paper aims at using various machine-learning algorithms to investigate the relationship between the metrics of urban features and crime occurrence. To do so we first architected datasets with both urban open data and features decomposed from street images. Then, several classification techniques such as Support Vector Machine, Decision Tree Classifier, Random Forest Classifier, and Gaussian Naïve Bayesian were applied to quantify the impact of demographic and street-view factors on street-related felony crime in New York City. Our results indicate over 60% accuracy on crime prediction. This study is valuable for understanding urban crime and for urban planner to design the urban perception.