Performance Evaluation of Intrusion Detection Systems on the TON_IoT Datasets Using a Feature Selection Method

As Internet of Things (IoT) technology develops so quickly, security issues with IoT devices have come to light. IoT is an array of intelligent devices connected via a network to provide various services. The amount of data generated by these devices has an impact on how well the current intrusion detection systems (IDS) function. The generated dataset consists of irrelevant features which reduces the performance of IDS, making IoT ecosystem vulnerable to cyberattacks. The researchers have suggested the feature reduction technique as a potential solution to the current problem. The proposed method seeks to reduce the feature count by removing the redundant feature subset. Several machine learning methods have been successfully implemented in this discipline. This study proposed the application of hybrid feature reduction technique. The research combined Convolutional neural network (CNN) and Long short-term memory (LSTM); CNN extracted local features and decreased dimensionality, while LSTM identified long-term relationships in the data. SVM and Random Forest classifiers models were used to classify the chosen feature subset. This study employed the TON_IoT Datasets, an up-to-date dataset, to evaluate the model. During data preprocessing, the study applied SMOTETomek data pre-processing technique to address class imbalance in the dataset. With the decreased feature subset, the classification models fared reasonably well; RF had a 98% accuracy rate while SVM had a 91% accuracy rate, showcasing the suggested methodology's potential for creating efficient IDS.