Crime Patterns and Prediction: A Data Mining and Machine Learning Approach

Abstract

Studying and analysing patterns in crime is of paramount importance in today’s world. With the increase of rapes, burglary, kidnapping and theft we need to provide a comprehensive framework for the government and law-makers for planning and informed decision making to control the increase of a particular kind of crime in various locations. Again, location and time of a crime have huge effects on the severity of the crime. According to a report published by the National Crime Records Bureau, which noted the crime rates between 1953 and 2006, the number of house burglaries in the country had dropped by 79.84% over a period of 53 years. However, the number of kidnapping cases in the country increased by 47.80% during that time. In addition to that, the total number of cognisable crimes under the Indian Penal Code (IPC) had shown a 1.5% increase in its numbers from 2005 to 2006. Looking at these statistics, we can understand the duplicity of crime data and how it changes over the years. Because of its dynamic nature, we need to find patterns in a crime which will help the police in the process. Multiple datasets were selected from government websites, which were used to find a pattern in the different classes of crime occurring in different states of the country. The dataset contains instances of reported crimes ranging from the year 1993-2014. With this information, we plan on predicting the crime rates in the future years using various machine learning algorithms and decide which algorithm is providing the most accurate values. Prediction of crimes can help the individual state police departments to concentrate their efforts more in the regions which recorded a higher concentration of crime or which shows a steady increase in its cases of reported crimes. 21 years of data is being used for training the model and extrapolating future values. This research aims at providing the people with an almost accurate prediction of the total number of crime instances in a State/UT within a span of 10 years from the last recorded year of data.