A Data Mining Based Approach For Burglary Crime Rate Prediction
Alghamdi, Dhaifallah M
MetadataShow full item record
Data mining techniques have been proven to be effective in many fields with the valuable information they provide. Those techniques can be deployed to lower crime rate by finding the patterns and links between the vast available data. Predictive policing is one of the areas that extensively depends on manual data analysis while it can benefit from these powerful techniques. Burglary is one of the most common crimes in Chicago and caused a loss of millions of dollars. The challenge in burglary prediction modeling is to develop a model that estimates the burglary rate based on a micro geographical level and to identify the contributing factors. In this thesis, data was collected from the city of Chicago data portal and the National Historical Geographic Information System (NHGIS). In addition to the demographic data, new data related to house characteristics and crime history was explored to study its effects on burglary rate. Based on these three categories, four experiment setups were designed to monitor the impact of each one on the model’s accuracy. A model which is based on micro geographical level, block groups, was developed to predict the burglary crime rate in Chicago. Four algorithms were compared and the Random Forest based model was the one with the highest accuracy. A time series approach was also used to develop a prediction model by using the moving average concept. However, the Random Forest based model was more accurate. Crime history data was found to have the biggest impact on the model outcome, while house characteristics data affected the model moderately. Sensitivity analysis and crime reduction strategies were included in this study for future planning. The proposed model addressed the challenge, estimated the burglary rate accurately and identified the contributing factors.