The Evaluation of Feature Selection Using Machine Learning for Cyber Attack Detection in Smart Grid

Smart grids have increased the performance and efficiency of power generation and distribution without any additional resources and manpower by implementing automation and IoT technologies. However, they are exposed to more aggressive cyber-attacks that use sophisticated techniques. The traditional IDS methods are predominantly either rule-based or signature-based approaches which will not recognize unknown threats. Therefore, it appears necessary to develop intelligent and adaptable IDS. Generally, To give a complete review of the feature selection techniques in machine learning based methods to detect cybersecurity attacks in smart grids. Feature selection is important in the cybersecurity attack detection in smart grids field because it is difficult to parse through analyzing large dimensionality data sets, causing redundancy and noise. The purpose of feature selection is to improve data with no or little information loss when removing relevance features or attributes, at the same time decrease the detection time and provide accuracy and interpretability. Case studies are provided for the number of features used in selected optimizing techniques; subsequently, the methods are categorized in three common techniques that encompass embedded; filtering & wrapper methods. Among analyzed studies, performance metrics used to compare primary selection were recall; accuracy and precision computational reliability. The case studies demonstrate that combining Random Forest with Support Vector Machines demonstrates promising detection in identifying attacks on a smart grid. The conclusions in this review showcase that optimized feature selection increases the machine learning capability and usability of resilient, real-time, scalable intrusion detection frameworks for critical digital infrastructures.