Feature Selection in Cyber-Attack Detection for Smart Grids Using Machine Learning Techniques and Random Forest

Abstract

The full review is now the application of feature selection techniques in machine learning in assessing cyber-attack detection by infrastructure within smart grids. Smart grids are currently becoming automated, communicating, and power-optimized via IoT technologies in energy generation, distribution, and consumption, therefore increasing exposure to cyber vulnerabilities. Feature selection from high-dimensional data becomes critical to building machine learning models, which are efficient and accurate in address timely detection and mitigation of cyber threats. Such an exhaustive review will allow feature selection strategies into three general classes: filter methods, wrapper methods, embedded methods; examine their effect concerning critical performance metrics: accuracy, precision, recall, and computational efficiency; and integrate the findings from recent studies to compare the effectiveness of these techniques for different attack scenarios such as denial-of-service, data injection, and false data manipulation. This translates into identifying trends, challenges, and gaps in contemporary methodologies that provide actionable recommendations and future directions in research aimed at building cyber infrastructures that are more resilient for smart grids and future research.