Enhancing Intrusion Detection with Advanced Feature Extraction Through Machine Learning and Deep Learning Methods

In this paper, the effectiveness of various ML and DL algorithms was investigated on the BotNeTloT-L01 dataset obtained from Kaggle., a comprehensive collection of IoT device traffic data. Initially., five ML classifiers., namely Random Forest, XGBoost, Decision Tree, Logistic Regression, and Extra Trees Classifier, along with an Artificial Neural Network (ANN), were employed to perform binary classification tasks on the dataset using all available features. Subsequently, feature extraction techniques, including ANOVA F-Test, Mutual Information, Recursive Feature Elimination, and Permutation Importance were explored to identify the best features in the dataset. After feature extraction, the same ML and DL algorithms were reapplied, and their performance was evaluated. Results demonstrated that feature extraction techniques significantly enhanced classification accuracy across all models, indicating the importance of feature selection in improving predictive performance. This work contributes to the advancement of IoT device security by providing insights into effective classification methods and feature selection strategies for detecting botnet attacks in IoT networks.