A Lightweight Intrusion Detection Method for Industrial Internet Based on Hybrid Feature Selection
Xiangdong Hu, Hao Nie · 2023
The deep integration of industrial control systems and internet technology has urged technological reformation, but with serious information security risks together; The industrial internet faces problems such as difficulty in intrusion detection, low detection accuracy, and high false alarm rates due to various data types and large redundancy. A lightweight intrusion detection method for industrial internet based on hybrid feature selection is proposed. Firstly, use information gain and Pearson correlation coefficient filtering to remove a large number of features unrelated to industrial internet security, and strengthen the correlation between selected features and security categories, and reduce the search interval and computational complexity for the subsequent wrapper method; Then, a sequence backward search algorithm is combined with an extreme learning machine to further select the optimal subset of malicious behavior features, improving the recognition of malicious behavior classification of the features; By using two-stage hybrid feature selection, data dimensionality reduction can be achieved to quickly identify malicious behavior, reduce computational complexity, and improve the accuracy and efficiency of industrial internet intrusion detection. The experimental results based on the natural gas pipeline dataset show that the proposed method has significant feature dimensionality reduction effect, with a 69.2% reduction in the number of core malicious behavior features compared to the original dataset. Moreover, the detection accuracy and recall rate based on the selected feature subset are increased by an average of 1.42% and 3.48% compared to the comparison method, and the detection time is shortened by an average of 0.14 seconds.