Intelligent Under Sampling Based Ensemble Techniques for Cyber-Physical Systems in Smart Cities

Abstract

Cyber-Physical Systems (CPSs) represent the next evolution of engineered systems that seamlessly blend computational and physical processes. The rise of technologies has brought about a heightened focus on security, making it a noteworthy concern. An intelligent ML-based CPS plays a pivotal role in analysing network activity within the CPS by leveraging historical data. This enhances intelligent decision-making to safeguard against potential threats from malicious hackers. The inherent uncertainties in the physical environment, CPS increasingly depend on ML algorithms capable of acquiring and leveraging knowledge from historical data to enhance intelligent decision-making. Due to limitations in resources and the complexity of algorithms, conventional ML-based CPSs face challenges when employed for operational detection in the critical infrastructures of smart cities. A lightweight intelligent CPS that is optimal, inexpensive, and can minimise the loss function is required. The widespread adoption of high-resolution sensors results in the presence of datasets with high dimensions and class imbalance in numerous CPS. Under-sampling-based ensemble algorithms ensures a better-equipped process to handle the challenges associated with imbalanced data distributions. The under-sampling-based ensemble technique solves class imbalance by lowering the majority class and establishing a balanced training set. This strategy improves minority class performance while reducing bias towards the majority class. The experimental findings validate the effectiveness of the proposed strategy in bolstering the security of the CPS environment. An assessment conducted on the MSCA benchmark IDS dataset affirms the promise of this approach. Moreover, the suggested method surpasses conventional accuracy metrics, striking a favourable balance between efficacy and efficiency.