Efficient Binary Intrusion Detection Using a Classifier-Aware Hybrid PSO–GA Feature Selection Framework

In recent times cyber-attack cases have risen at a much faster pace compared to previous decades. Intrusion detection systems are one of the key systems for out covering modern networks from cyberattacks. The two aspects which always come into notice are complexity and computational cost which is proportional to high-dimensional traffic data. This research work emphasizes on the applying binary intrusion detection framework in efficient way to overcome cyber-attacks. Here the framework enables classifier aware hybrid PSO–GA feature selection techniques by utilizing CICIDS2017 dataset. The flow of the work stated with preprocessing to ensure the data is set in ideal way followed by mutual information filtering. Later the wrapper enabled hybrid optimization was introduced by considering local search refinement. This refinement not just detects but also gives a concise information about all the considered subsets of features. Shortlisted features were treated with models such as Random Forest, LightGBM, XGBoost, and Multilayer Perceptron. A detailed comparison is projected for before and after feature selection are also reported. Interestedly, generate results give clear information on the reduction of the feature set from 70 to 5. This sudden reduction is key aspect of this work. Apart from reduction the performance was not compromised for tree-based models. Specifically, the behavior of LightGBM, Random Forest, and XGBoost consistently gave high accuracy results. While F1-scores balanced the precision efficiently. With consolidated outcome the reported framework not just effectively balanced the detection accuracy but also focused on the computational efficiency in smoother way by making it more practical approach in intrusion detection applications.