Lightweight Malware Detection In Smartphones Through Predictive Analytics, Feature Engineering, And Hybrid Optimization Technique

Abstract

Android smartphones are rapidly growing in number and the proliferation of these devices has also led to the development and deployment of sophisticated malware that is capable of bypassing traditional signature-based detection methods. The mobile environment's resource limitations require detection systems to be both accurate and lightweight. In this paper, a lightweight hybrid malware detection system for smartphones using feature engineering, predictive analytics, and data mining technique with two complementary optimization algorithms is proposed: Grey Wolf Optimizer (GWO) algorithm for intelligent feature selection and Mutual Information-based Feature Ranking (MIFR) as a deterministic mathematical baseline. The proposed pipeline starts by extracting a detailed feature set for Android application packages that includes permissions, API calls, system calls and intent patterns. GWO reduces the space of the features iteratively by simulating the predator-prey hunting behavior, whereas MIFR gives an analytical ranking to ensure the interpretability. The optimized feature subsets are then used as inputs to a Random Forest classifier which has an accuracy of 98.6%, a precision of 98.2%, a false-positive rate of 0.9%, a dimensionality reduction of 86.3%, and reduced the number of features from 342 to 47. The complexity analysis demonstrates that GWO-MIFR is in O(n log n) amortized, which is appropriate for on-device inference. The comparative assessment with the most up-to-date methods in the field proves that the proposed system achieves consistent superior accuracy and computational costs compared to the current methods. The findings show that hybrid optimization-driven feature engineering is a viable approach for mobile malware detection in production.