DKRF: Machine Learning with Optimised Feature Selection for Intrusion Detection

The rapid development of IoT (Internet of Things) devices affords intruders to launch many attacks resulting in a security breach. Thus, a network IDS (Intrusion Detection System) is vital that assists in monitoring the network traffic for automatic intimation of abnormal events through alerts. It is also essential for securing a network as it permits detecting and responding to malicious traffic. The main benefit of IDS is to confirm that IT personnel is warned when any network intrusion or attacks occur. Though conventional research has attempted to achieve a better intrusion detection system, it lacked with respect for effective detection rate resulting in a higher false alarm rate. To solve this, the present study considers ML (Machine Learning) based methods as they can boost the accuracy and robustness of the system. Initially, the IoT data is clustered and fed into the proposed Optimised PSOGA (Particle Swarm Optimisation Genetic Algorithm) for feature selection. Its ability to search large spaces and better representation with effective data management having numerous features has made it suitable for selecting only the relevant features. Further, classification is performed, by the introduced DKRF (Dynamic K-means Random Forest), due to its assured convergence and flexibility in solving classification issues. These advantages, makes the proposed method to possess effective performance in detecting intrusions. This effectiveness is measured based on accuracy, FPR (False Positive Rate), precision, F-measure, and recall.