Abstract
Crime analysis has become a critical area for helping law enforcement agencies to protect civilians. As a result of a rapidly increasing population, crime rates have increased dramatically, and appropriate analysis has become a time-consuming effort. Text mining is an effective tool that may help to solve this problem to classify crimes in effective manner. The proposed system aims to detect and classify crimes in Twitter posts that written in the Arabic language, one of the most widespread languages today. In this paper, classification techniques are used to detect crimes and identify their nature by different classification algorithms. The experiments evaluate different algorithms, such as SVM, DT, CNB, and KNN, in terms of accuracy and speed in the crime domain. Also, different features extraction techniques are evaluated, including root-based stemming, light stemming, n-gram. The experiments revealed the superiority of n-gram over other techniques. Specifically, the results indicate the superiority of SVM with trigram over other classifiers, with a 91.55% accuracy.