Paper Title
A Predicting and Adapting Malware Sustainability Using Hybrid Temporal Behavior Fusion Framework

Abstract
Accurate, flexible, and computationally sustainable detection systems are required due to the swift spread of Android applications and the growing complexity of mobile malware. Current retraining-based and incremental learning techniques find it difficult to strike a balance between long-term adaptability, efficiency, and detection accuracy in the face of sudden structural changes in the Android SDK and evolving concept drift. The Hybrid Temporal Behavior Fusion Framework proposed in this paper consists of three integrated layers: a Behavioral Ensemble Detection Module that uses year-adaptive Random Forest classifiers with Attention-Based Feature Weighting, a Temporal Inference and Clustering Module that uses DBSCAN and Gradient Boosted Decision Trees for probabilistic release-year estimation, and a Drift-Aware Optimization Module that uses Kullback–Leibler Divergence for selective recalibration without complete retraining. The suggested framework ensures reliable and scalable malware detection in real-world evolving threat environments by achieving detection accuracy and F1 scores comparable to retraining-based methods while maintaining significantly lower computational overhead, as tested on an Android dataset spanning 2014–2023 Keywords - Android Malware Detection, Concept Drift Adaptation, Temporal Inference and Clustering ,Hybrid Ensemble Learning ,Kullback Leibler Divergence ,Sustainable Machine Learning.