Nature-inspired based ensemble feature selection and stacked ensemble classifier fusion for Android malware detection

Android is widely used for tablets and smartphones. Thus, Android app malware has grown quickly in recent years. Malware detection in these apps is effective with ML algorithms. A reliable and effective malware detection approach is still difficult due to the vast number of features and high-dimensional dataset that provides the lowest accuracy despite research and industry efforts. Feature selection involves finding and removing features from a dataset while retaining class label variance. Using dynamic analysis of Android malware samples, this study introduces the nature-inspired ensemble feature selection (NIEFS) and stacked ensemble classifier fusion (SECF) multi-classification models. The NIEFS model uses evolutionary computation methods like Fuzzy Membership Grasshopper Optimisation Algorithm (FMGOA), Lévy flight pigeon-inspired optimisation (LEFPIO), and Cauchy Operator Squirrel Search Algorithm (COSSA) to remove redundant or irrelevant features and select relevant ones to improve detection accuracy. Multilayer SECF integrates EC-created malware outputs using a Mutual Information (MI)-based ensemble approach, which has good detection accuracy. Machine learning methods like J48, mean weight deep belief network (MWDBN), REPTree, and Voted Perceptron can be combined. Finally, classifier performance was tested using MATLABR2020a, Precision (Pre), Recall (Rec), F-measure (FM), and Weighted F-measure.