MLCFSM: Multi-Layer Common Feature Selection Methods for Robust Intrusion Detection in IoT Network

This paper examines the crucial role of feature selection (FS) methods in enhancing IDS models within IoT security frameworks. With IoT devices generating vast and diverse data, traditional IDS models face the problem of having too many dimensions or data feature space, which hinders their effectiveness. Utilizing FS methods becomes imperative to streamline datasets, improving model performance, and strengthening security measures. The proposed method combines the strengths of filter and wrapper methods to create a subset feature dataset. Initially, three filter methods, including fast correlation-based feature (FCBF), mutual information (MI), and ANOVA F-test (ANOVA), as well as one wrapper method, Forward Sequential FS (FSFS), are applied to the entire dataset to select important features by ranking. We identify the common features in the filter and wrapper methods by constructing an intersection subset. The final feature subset dataset is used to train the proposed MLCFSM (Multi-Layer Common Feature Selection Methods) using 10 classifier algorithms and evaluated on three datasets: FLNET2023, ACI-IoT2023, and CIC-IoT2023. The MLCFSM demonstrates significantly improved performance in terms of the evaluation metrics, achieving up to 99.999% mean accuracy and reduction in computation cost. By reducing dataset dimensionality through MLCFSM, our method demonstrates a competitive performance with state of the art even while using a small feature space.