Performance Evaluation of Machine Learning Algorithms for Spam Profile Detection on Twitter Using WEKA and RapidMiner

Twitter social network is growing on a daily basis and as a result, attackers have developed interest in distributing malicious contents on this platform. Numerous studies have investigated the possibility of reducing spamming activities on Twitter with each study focusing on introducing a new set of features for countermeasure. This paper adopts the set of features for identifying spammers on Twitter and introduces additional features to improve classifier performance. The performance of four machine learning algorithms: Random forest (RF), Support vector machine (SVM), K nearest neighbor (KNN), and Multilayer perceptron (MLP) across two popular machine learning tools-WEKA and RapidMiner were evaluated. Results from the experiment show that SVM, KNN, and MLP on WEKA outperformed those algorithms on RapidMiner. However, in the case of RF, RapidMiner achieved higher accuracy compare to RF on WEKA. Based on the 32 features in the dataset, MLP and RF on both WEKA and RapidMiner outperformed other classifiers with accuracy of 95.42% and 95.44% respectively. These findings would be useful for researchers willing to develop a machine learning model to detect malicious activities on social network.