Hybrid momentum accelerated bat algorithm with GWO based optimization approach for spam classification

Spam emails have become more prevalent, necessitating the development of more effective and reliable anti-spam filters. Internet users face security threats, and youngsters are exposed to inappropriate content while receiving spam emails. The gigantic data flow between billions of people and the tremendous number of features (attributes) makes the task more tiresome and complex. Feature Selection (FS) technique is essential for overwhelming accuracy, time and spatial complexity when we have high dimensional data (i.e., the number of features is very large). Spam emails have been successfully filtered and detected using Machine Learning (ML) methods by various researchers nowadays. This work proposes a hybrid binary Metaheuristic Algorithm (MA) based Feature Selection (FS) approach for classifying email spam. The proposed FS approach is based upon two MA, i.e., Bat Algorithm (BA) with Grey Wolf Optimization(GWO). A novel concept of bat momentum has been introduced here, replacing the previous bat velocity. Two quantity, i.e., velocity and momentum, has an entirely different effect on the particle (i.e. bats). But they always follow the exact directions for both of them. To provide the best possible set of features for the FS process, the proposed approach uses an amalgamation technique to reach both the global and local optimum solution. To get the global optimum solution, a new momentum-based equation has been added to the BA, substituting the velocity equation from the prior BA. The GWO property has been added to the momentum-based equation mentioned above to improve the FS process search capabilities. Here a novel concept convergence timer has been introduced, which can eliminate the convergence issue in the iterative algorithm if it arises. A novel GWO based lévy flight update has been introduced here to produce the local optimum solution. We have evaluated our proposed method on two benchmark spam corpora (Spambase, SpamAssassin) having different significant properties. The proposed FS approach has been tested on various classification and clustering algorithms to check the robustness and how the model will behave on unknown data. After comparing multiple state-of-the-art and existing approaches, the proposed method is superior in boosting classification accuracy while minimizing the features in the feature set for misclassifying legitimate emails as spam.