Ensemble Learning Based Feature Selection for Detection of Spam in the Twitter Network

Abstract

The proliferation of spam in Twitter poses serious social problems which affects society by causing people to commit crimes. For a better spam detection model, the challenge is to select optimal lightweight features that should be feasible to process a large number of tweets in very less time. This chapter focuses on discovering the optimal features from the dataset, developing various machine learning models for detection of spam and finally checking the stability of machine learning algorithms for varying sizes of training data. The dataset consists of user account and tweet based features from which the prominent features are selected using ensemble feature selection by employing different feature selection algorithms. Further, the effect of prominent features for spam detection is investigated using various machine learning algorithms. Comparative analysis of spam detection models is performed with the objective of finding the ones with the best performance and stability on large data sets of Twitter. The experimental results show that random forest classifiers perform better with an accuracy of 86.95%