ADAPTIVE MULTI-OBJECTIVE FEATURE OPTIMIZATION USING RECURSIVE MONTE CARLO TREE SEARCH FOR NETWORK INTRUSION DETECTIONS

Abstract

Intrusion detection in modern network environments is challenged by high-dimensional data, imbalanced attack classes, and constantly evolving cyber threats. Traditional intrusion detection methods often rely on fixed or single-objective feature selection techniques, limiting their ability to achieve high detection performance while maintaining computational efficiency. This paper proposes an adaptive feature optimization framework that combines recursive Monte Carlo Tree Search (MCTS) with multi-objective optimization for effective intrusion detection. The proposed Adaptive Recursive Multi-Objective Feature Selection (ARMFS) framework progressively refines feature subsets through recursive search, enabling efficient exploration of the feature space. Feature subsets are evaluated using two objectives: maximizing classification accuracy and minimizing the number of selected features. This strategy identifies compact and informative feature sets while reducing computational complexity. The optimized features are used to train a deep neural network classifier, with its hyperparameters tuned using a nature-inspired optimization algorithm to improve convergence and generalization. Experimental results on benchmark intrusion detection datasets demonstrate that the proposed framework achieves high detection accuracy with significantly fewer features than the original datasets. The method also improves the detection of minority attack classes, maintains low false positive rates, and produces consistent performance across multiple validation runs. These results demonstrate that combining recursive search with multi-objective feature optimization provides an efficient and scalable solution for intelligent intrusion detection in dynamic and high-dimensional cybersecurity environments.