Securing Network: Integrating Machine Learning Approaches for Intrusion Detection System

In this modern era, everyone is connected through network systems. Its remains a greater challenge to protect any organizational network system as very confidential and sensitive data, resources, and assets are integrated with these systems. Regarding this, intrusion detection and prevention is a major research domain to secure any network infrastructure to mitigate miscellaneous cyber-attacks. This research aims to investigate the pros and cons of intrusion detection systems, and create a sustainable model. In this approach, we have deeply discussed several both supervised and unsupervised machine learning algorithms on updated UNSW-15 dataset. To establish an effective model, we conducted our experiments on Random Forest, AdaBoost, K-Nearest Neighbors (KNN), Decision Tree (DT), and GRU (Gated Recurrent Unit) algorithms. As the dataset contains a high dimensionality of features, feature selection techniques using Heatmap and co-relation process have been applied to remove mostly co-related features. To obtain the best results, we split the dataset into 70 percent as a training dataset and 30 percent as a test dataset. Based on the experimental results, it concluded that Random Forest algorithms outperformed all other classifiers by achieving a high accuracy of 95% and other parameters like precision, recall, specificity, F1-score, and roc-auc score. In future, more algorithms, especially deep neural network-based algorithms will be integrated to create ML model with higher accuracy rate for the detection and classification of intrusion detection.