Computer Science · Computer Networks and Communications
Machine Learning-Based Feature Selection Intrusion Detection System for Enhancing IoT Network Security
Vivek Kadam, Bhushan S. Deore, Jotiram Krishna Deshmukh, Suresh Pingale, Vijayshri Khedkar (24218643), Shamal Salunkhe
See on the mapAbstract
The rapid proliferation of Internet of Things (IoT) devices has introduced unprecedented connectivity but simultaneously amplified cyber-security risks.The IoT networks are vulnerable to breaches and cyber attacks because of compromised security protocols and inadequately secured IoT devices. These devices have constrain over security update and lack of encryption make them prone to cyber attacks. This paper presents a comprehensive study on developing a robust intrusion detection system for IoT networks by leveraging an ensemble machine learning approach. The methodology encompasses a complete pipeline from raw traffic capture and sophisticated feature extraction using CICFlowMeter, through meticulous data pre-processing to advanced feature selection and the implementation of diverse machine learning models. Further it presents a strategic machine learning-based IDS framework using a stack classifier incorporating optimized feature selection to enhance detection accuracy and efficiency. Modelling using the K-Best algorithm to select features and extract the top 15 most important ones and building an ensemble model are the two mainstays of this method. Classification algorithms including Random Forest (RF), Naive Bayes (NB), Support Vector Machine (SVM), K-Nearest Neighbors (KNN) are compared with proposed custom-built Stack Classifier by considering the ‘Ton IoT dataset’. The findings unequivocally demonstrate the superior performance of the Stack Classifier, achieving significantly higher accuracy, precision, recall, and F1-score, thereby identifying it as the optimal model for enhanced intrusion detection.