Crime Prediction Using Twitter Data

Abstract

As social media platforms are being increasingly used across the world, there are many prospects to using the data for prediction and analysis. In the Twitter platform, there are discussions about any events, passions, and many more topics. All these discussions are publicly available. This makes Twitter the ultimate source to use the data as an augmentation for the decision support systems. In this paper, the use of GPS tagged tweets for crime prediction is researched. The Twitter data is collected from Chicago and cleaned, and topic modelling is applied to the resultant set. Before topic modelling, an algorithm has been developed to identify tweets that are relevant to the crime prediction problem. Once the relevant tweets are identified, topic modelling is applied to find out the major crimes in the different beats of Chicago. Kernel density estimation (KDE) is applied to traditional data. The result of this and topic modelling are used to predict the crime count for each beat using logistic regression.