Feature Sequencing Method of Industrial Control Data Set Based on Multidimensional Evaluation Parameters.

Area of Science:

• Industrial Control Systems Security
• Machine Learning for Anomaly Detection
• Data Science and Feature Engineering

Background:

• Industrial control datasets often contain numerous features with high redundancy, negatively impacting neural network training speed and anomaly detection accuracy.
• Traditional dimension reduction techniques can increase false positive and false negative rates due to feature independence.
• Feature sequencing algorithms offer a way to mitigate these negative effects.

Purpose of the Study:

• To propose an adaptive feature sequencing method for selecting optimal algorithms based on dataset characteristics.
• To enhance the efficiency and accuracy of neural network-based anomaly detection in industrial control systems.
• To provide a systematic approach for algorithm selection tailored to diverse datasets.

Main Methods:

• An evaluation index system was constructed using dataset basic information, mathematical characteristics, and association degrees.
• A decision tree model was trained using dataset labels and evaluation indices to select the most suitable feature sequencing algorithm.
• Experiments were performed on 11 diverse datasets, including Batadal, CICIDS 2017, and Mississippi.

Main Results:

• The adaptive method demonstrated improved performance across multiple datasets.
• Classifying sequenced datasets using ResNet resulted in an average accuracy increase of 2.568% over 30 generations.
• The average reduction in training time per epoch was 24.143%.

Conclusions:

• The proposed adaptive feature sequencing method effectively selects algorithms for optimal comprehensive performance in anomaly detection.
• This approach addresses the challenges posed by high-dimensional and redundant industrial control datasets.
• The findings suggest a significant improvement in both the accuracy and efficiency of anomaly detection systems.