A Framework to Predict Social Crime through Twitter Tweets By Using Machine Learning

An increasing amount of data and information coming from social networks that can be used to generate a variety of data patterns for different types of investigation such as human social behavior, system security, criminology etc. A framework is developed to predict major types of social media crimes (Cyber stalking, Cyber bullying, Cyber Hacking, Cyber Harassment, and Cyber Scam) using the data obtained from social media website. The proposed framework is consist of three modules; data (tweet) pre-processing, classifying model builder and prediction. To build the prediction model Multinomial Naive Bayes (MNB), K-Nearest Neighbors (KNN) and Support Vector Machine (SVM) is used that classify given data into different classes of crime. Further N-Gram language model is used with these machine learning algorithms to identify the best value of n and measure the accuracy of the system at different levels such as Unigram, Bigram, Trigram, and 4-gram. Results shows that all three algorithm attain the precision, Recall and F-measure above than 0.9 however Support vector machine performed slightly better. The proposed system produced better accuracy result as compared to existing network-based feature selection approach.