Optimal Ensemble Learning with Meta-heuristics for Multiclass Classification of Syscall-Binder Interactions in Mobile Applications
The paper elaborates on the relation between syscalls and a Binder mechanism. System calls are a type of instruction that enables applications to converse with the core, whereas Binder mechanisms are those through which applications and services interact with each other. Due to the rapid growth in mobile application usage, it becomes crucial to understand such interactions and take necessary actions to prevent potential attacks. These are ensemble learning methods. One of the techniques from the set of machine learning methods that involve the combination of more than one model, combined with optimization strategies, is known as hyper-metaheuristics. We conducted an experiment using three machine learning models: GBM, RF, and DT, each improved by hyper metaheuristics, which achieved an accuracy rate of 99.18%, 98.88%, and 99.70%, respectively. Other important metrics, such as precision, recall, and F1-score, were also exceptionally well-performed by these models, proving efficient in detecting potential security threats. In general, this research proposes a novel but efficient approach toward identifying security vulnerabilities in mobile applications and contributes to safer mobile ecosystems in today's digital landscape.