Research on Abnormal Traffic Detection of Internet of Things Based on Feature Selection

The Internet of Things (IoT) is under attack with the spread of IoT device applications. Traffic anomaly detection is critical to ensure IoT security. This paper proposes a method for abnormal traffic detection in IoT based on random forest and cascade deep learning detection model. IoT device traffic data were analyzed by random forest algorithm to screen out feature parameters with significant relevance to the detection target. To detect abnormal traffic, the feature parameters were taken as the input of convolution neural network (CNN) and long short-term memory (LSTM) detection models and to compare and analyze different dimensions, classification methods, and algorithmic model approaches. The results show that the detection accuracy reached 99.98% in dichotomous classification and 88.14% in multi-classification for high-dimensional data. The proposed CNN and LSTM cascade model detection methods were more stable than CNN and LSTM methods. Additionally, the model solved the problem of high dimensionality and nonlinearity of a large number of parameters to ensure that the main characteristics of the input variables were remembered over time. Therefore, the detection accuracy of this hybrid model is more desirable than that of the CNN and LSTM models without feature selection.