Feature selection using chi square to improve attack detection classification in IoT network: work in progress

Abstract

To maintain network security, Intrusion Detection System (IDS) is needed to detect anomaly and attack. Designing proper IDS requires accurate model. This paper proposes a model, which consists of statistical extraction, feature selection, dataset clustering, classification, and performance measurement. Experiments on MQTT-IOT-IDS2020 dataset which contains Normal, scan A, scans U, Sparta and mqttbruteforce are conducted. The dataset is statistically extracted using Bidirectional-based features packet header feature with 37 features. Chi square algorithm is selected for performing feature extraction process. 10 relevant and best features are selected and ranked into 5-subsets and 10-subset feature. Three dataset splitting into testing data and training data of 90%:10%, 70%:30% and 50%:50% are created. Binary classification using k-Nearest Neighbor (KNN) and Adaboost algorithms are performed. The experimental results show accuracy level above 99% for all scenarios, with Adaboost algorithm outperforms k-Nearest Neighbor algorithm.