DDoS Detection in Smart Home IoT Networks: An Experimental Comparison of MLP and AE-MLP with Correlation-Based Feature Selection

Abstract

As the use of the Internet of Things (IoT) increases, the number of smart homes is rapidly increasing. But the increase in the number of connected devices increases the attack surface, making them vulnerable to Distributed Denial-of-Service (DDoS) attacks. The Autoencoder-Multilayer Perceptron (AE-MLP) is a hybrid deep learning model for DDoS attack detection in the smart home context. The Multilayer Perceptron (MLP) is proposed as the baseline model. The dataset is collected from a testbed of several IoT devices deployed in smart homes. Flow based features were extracted from network traffic in PCAP format using the CICFlowMeter. Data cleaning, feature selection with the correlation matrix and Min-Max normalization were the steps of data preparation process. We tried different data partition, optimizer, learning rate and epoch settings. Results showed that the performance of both models was greatly improved with feature selection. Before feature selection, the accuracy and F1-Score of the AE-MLP are 99.2% and 99.3% whereas the accuracy and F1-Score of the MLP are 99.5% and approximately 68%. The results indicate that feature selection plays a significant role in reducing noise and enhancing the performance of DDoS detection in smart home environments.