Android Malware Detection (AMD) Using Stacked Deep Learning Ensemble Classifier Fusion (SDLECF) with Nature-Inspired Based Ensemble Feature Selection (NIEFS)

Android is the most rapidly expanding mobile computer platform, which has been targeted by a variety of malware. It can be effectively identified using Deep Learning (DL) techniques. Typical feature selection algorithms disregard feature correlation which has been solved by using wrapper-based feature selection models. Wrapper-based techniques take a lot of time to select feature subsets. In this paper, Nature-Inspired Based Ensemble Feature Selection (NIEFS), and Stacked Deep Learning Ensemble Classifier Fusion (SDLECF) classifier has been introduced for combination of DL methods with increased detection accuracy. NIEFS model is developed based on a variety of evolutionary computation approaches, such as the Cauchy Operator Squirrel Search Algorithm (COSSA), Lévy Flight Pigeon-Inspired Optimization (LEFPIO), and Fuzzy Membership Grasshopper Optimization Algorithm (FMGOA) for eliminating redundant or unnecessary features. The outputs of various approaches have been integrated using Mutual Information (MI). SDLECF is introduced by merging many models (Bidirectional Gated Recurrent Unit (Bi-GRU), Sparse Autoencoder based Deep Neural Network (SAE-DNN), Bidirectional Long Short-Term Memory (BDLSTM), and Mean Weight Deep Belief Network (MWDBN)) to attain highest malware detection performance. Bi-GRU can handle data sequences in both forward and backward direction. SAE-DNN includes of three components like an encoder, a decoder, and a classification. BDLSTM classifier is a category of Recurrent Neural Network (RNN) which works on both forward and backward directions. MWDBN includes of Multiple Restricted Boltzmann Machine (RBM) layers for classification. Finally, classifier performance was measured using MATrix LABoratory R2020a (MATLABR2020a) and the metrics like Precision (Pre), Recall (Rec), F-measure (FM), and Weighted F-measure (WFM).