Impact of Data Balancing and Feature Selection on Machine Learning-based Network Intrusion Detection

Unbalanced datasets are a common problem in supervised machine learning. It leads to deeper understanding of the majority classes Therefore, learning model is more effective at recognizing than minority classes. Naturally, imbalanced data, such as disease data and networking, has emerged real life. DDOS one network intrusions found happen often R2L. There an imbalance composition attacks Intrusion Detection System (IDS) public NSL-KDD UNSW-NB15. Besides, researchers propose many techniques transform it into balanced by duplicating class producing synthetic data. Synthetic Minority Oversampling Technique (SMOTE) Adaptive (ADASYN) algorithms duplicate construct for Meanwhile, can capture labeled data's pattern considering input features. Unfortunately, not all features have equal impact on output (predicted or value). Some interrelated misleading. important should be selected produce good model. In this research, we implement recursive feature elimination (RFE) technique select from available dataset. According experiment, SMOTE provides better dataset ADASYN UNSW-B15 with high level imbalance. RFE selection slightly reduces model's accuracy but improves training speed. Then, Decision Tree classifier consistently achieves recognition rate Random Forest KNN.