Machine Learning Techniques Evaluation with SMOTE on IoT-23 Dataset

Machine Learning (ML) algorithms become one of the most significant techniques for building intrusion detection systems (IDS)s. The most important part of the ML based IDS is training phase, which depends on the quality of the dataset that fed to the model. The important problem that challenges most ML techniques is using imbalance datasets for training. This work investigates the impact of the imbalanced dataset on the accuracy rate of three ML techniques (KNN, SVM and ANN). The work uses SMOTE technique for eliminating such imbalances in a dataset. The work evaluates the accuracy rate of three major ML algorithms with Synthetic Minority Oversampling Technique (SMOTE) method using an up-to-date dataset that used for IoT based IDS. Results show that SMOTE can improve the accuracy of all three ML based IDS, however, differently.