Enhancing the Efficacy of Machine Learning-Based Intrusion Detection Systems on an Imbalanced and Contemporary Dataset

As the Internet has increased in popularity, more computers are networked and becoming part of our daily lives. Common and less evident server vulnerabilities may be used by hackers to access systems and conduct complicated assaults. An Intrusion Detection System (IDs), a popular computer security tool, uses machine learning on a previously acquired dataset. The datasets were generated in distinct networks across brief periods and may not contain the latest information. They're bloated and can't hold enough data to survive an attack. Modern intrusion detection systems are less effective with inconsistent and outdated information for rare attack paths. This paper presents five machine-learning intrusion detection solutions. K Nearest Neighbor (KNN), Decision Tree (DT), Gradient Boosting (GB), Adaboost (AB), and Random Forest (RF) are used. The latest security dataset CSE-CIC-IDS2018 replaces earlier, more often-used datasets to improve Intrusion Detection Systems (IDS). An unbalanced dataset was selected. The imbalance ratio is reduced using the Synthetic Minority Oversampling Technique (SMOTE), a synthetic data model. The approach increases attack type-specific system performance while lowering false alarms and missing incursions. Minor classes increase their data amounts to reach the average data size. Experimental results demonstrate that the recommended strategy greatly enhances uncommon intrusion detection.