Towards Detection of DDoS Attacks in IoT with Optimal Features Selection

The exponential growth of internet-enabled devices and their interconnectedness heightens the vulnerability of technology to cyber threats. The simplicity of communication lures attackers to execute numerous attacks, with Distributed Denial of Service (DDoS) emerging as a major threat due to its challenging detectability. Over recent years, numerous machine learning mitigation methodologies have arisen to combat this issue. In this paper, we present an approach for detecting DDoS attacks, with a primary focus on optimal feature selection and data pre-processing to mitigate the risk of overfitting and enhance accuracy. We employ an embedded method utilizing a decision tree in Recursive Feature Elimination with Cross-Validation (RFECV) to select the most effective features. Subsequently, we apply Gradient Naïve Bayes (GNB), Decision Tree (DT), Random Forest (RF), and Binary Classification using deep neural network deep learning models. These models undergo validation using the CICDDoS2019 dataset. Performance evaluation reveals that the deep learning model surpasses others, achieving an accuracy of 99.72%.