Feature-Driven DDoS Detection in IoT with Machine Learning Approach

This research focuses on strengthening Internet of Things (IoT) security against Distributed Denial-of-Service (DDoS) attacks by implementing a hybrid machine learning approach that integrates Naïve Bayes, K-Nearest Neighbors (KNN), and LightGBM algorithms. Recognizing the critical vulnerabilities of IoT systems to DDoS attacks and the urgent need for robust security measures, the study evaluates these models using the CICIoT2023 dataset. To address challenges like data imbalance, sophisticated feature selection techniques, including correlation matrix and information gain, are employed. The research rigorously assesses the models' performance in terms of accuracy and false positive rates, aiming to determine the most effective method for mitigating cyber threats and enhancing IoT system resilience. In Split 1, KNN achieves 99.44% accuracy with a false positive rate of 0.05%, LGBM achieves 99.89% accuracy with a false positive rate of 0.04%, while NB lags with 88.67% accuracy and a false positive rate of 0.2%. In Split 2, KNN reaches 99.8% accuracy with a false positive rate of 0.01%, LGBM achieves 99.9% accuracy with no false positives, and NB maintains 89.69% accuracy with a false positive rate of 0.2%. These findings highlight the effectiveness of the machine learning approach in mitigating cyber threats and fortifying IoT system resilience.