An Enhanced AdaBoost-Based Cyber-Physical Attack Prediction Framework for Smart Grid Systems Using Advanced Data Balancing Techniques

Abstract

Cyber-physical security has become a critical challenge in smart grid (SG) systems due to the tight integration of digital communication with physical power infrastructures. This paper proposes a machine learning (ML)–based framework for the precise prediction of cyber-physical attacks using the “Smart Grid Cyber-Physical Attack Dataset” from IEEE Dataport, which contains synchronized cyber and physical measurements under both benign conditions and multiple attack scenarios, including ransomware, brute force, reverse shell, backdoor, and false data injection attacks. The methodology includes several preprocessing steps such as merging cyber–physical data sources, handling missing values, encoding categorical attributes, and standardizing numerical features. Given the severe class imbalance within the data set, a range of balancing techniques to improve class distribution and enhance model stability. AdaBoost was adopted as the primary classifier and was evaluated using accuracy, precision, recall, F1-score, AUC-ROC, and execution time. The experimental results demonstrate significant performance improvements, with ADASYN and ADASYN–Tomek Links achieving 91.55% accuracy with 94.88 F-score. These findings confirm the high efficiency of the balanced AdaBoost-based framework for detecting cyber-physical attacks in smart grid environments.