Hybrid Spam Filtering using Monarch Butterfly Optimization Algorithm with Self-Adaptive Population

Abstract

Spam causes bottlenecks and congestion, reducing the speed, processing power, available memory, and bandwidth. Existing spam email classification methods need to be more accurate because of the large dimensionality of hybrid spam datasets. This makes the need for a feature dimensionality reduction technique that uses only associated features of the problem instead of all features in the dataset. This paper presents a feature selection based on the monarch butterfly optimization (MBO) algorithm that emphasizes less complexity and few features. This method is efficient and produces a more accurate classification. To improve further standard MBO algorithm performance, we introduce the population size in both subpopulations 1 and 2 will experience dynamic variations as the algorithm proceeds along its linear way. As the idea of a self-adaptive and greedy strategy is modified, the self-adaptive population monarch butterfly optimization (SPMBO) method is introduced, and only newly generated SPMBO individuals are eligible for the next generations if they are better individuals earlier before. Later, this paper proposes an email classification system based on k-nearest neighbors (k-NN) based on two distance metrics, explicitly Euclidean, and Manhattan, that also uses the SPMBO technique. This method seeks to determine whether a hybrid email is a spam. The efficiency of the proposed SPMBO algorithm was compared with standard MBO based on three datasets Dredze, Image spam hunter, and Spambase. Thus, the use of SPMBO results has shown superior as related to other authors' works in relevant fields.