Parameter based Threat Detection in Network Datasets using Machine Learning algorithms

Abstract

Intrusions are consistent torment that networks have to encounter. To get rid of these malicious attacks a variety of approaches have been developed, the use of machine learning and deep learning is considered the most effective strategy to detect multifarious attacks. This research presents two different intrusion detection systems that have been trained by applying machine learning algorithms like Random Forest (RF), Support Vector Machine (SVM), and deep learning algorithm Long-Short Term Memory (LSTM). For better training of algorithms, we have developed an effective training dataset by merging BoT-IoT and NB-15 and balancing the classes. For machine learning using Random Forest (RF) and Support Vector Machine (SVM), we get the accuracy of 98.32% and 99.60% respectively. For deep learning using LSTM, we achieve an accuracy of 99.93% for 3-class classification and 99.98% for 5-class classification. Hyperparameter tuning and cross-validation techniques have also been used for effective training, and less model execution or training time. The results show that our approach of merging and balancing datasets outputs better accuracy as compared to state-of-the-art machine learning and deep learning algorithms