Crime analysis in India

Abstract

Crime records of India are available at a district-level on the National Crime Records Bureau website. We looked to build models which could explain the crime rate at a district-level because the explanatory variables show vast diversity at the state-level.To augment this, we searched for district-level predictor data, which we selected based on secondary research and domain knowledge. Typically, crime rate is affected by, or rather thought to have been affected by macroeconomic conditions prevalent in the region – population, literacy, sex ratio, level of urbanisation, etc. Since district-level data of these parameters is not readily available throughout all of the years, we decided to perform two types of analysis: Multiple Regression Models on cross-sectional data of 2001 (district-level) Panel Data Models on panel data of 2001 and 2011 (district-level) as key factors were available only for census years Census Data for the predictors was extracted and used from Open Government Data Platform, CMIE States of India database, IndiaStat Database and District Metrics Database. Preprocessing steps involved correcting mismatch of district names across different datasets used, normalization of factors with respect to population and outlier removal. Data visualization is also performed to provide easy high-level view of the key crime statistics prevailing in India at the state level.