Abnormal Traffic Detection in Data Mining Techniques using Scaled Exponential Linear Unit based Deep Autoencoders

With the advanced development of data mining techniques, the abnormal traffic detection is becoming simple and easy to acquire. The existing approaches are complex to estimate the time series data in parallel as well as vanishing or gradient explosion problems are most often occurring. Hence, this research proposes the Scaled Exponential Linear Unit based Deep Autoencoders (SELU-DAE) approach for the abnormal traffic detection according to the data mining techniques. Initially, this research collects the two public data sets such as NSL-KDD and UNSW-NB15 for the estimation of model performance. Then, the Min-Max normalization technique is utilized to equally scale the input features and then, the Principal Component Analysis (PCA) technique is utilized to select the significant features in the features selection process. Finally, the SELU-DAE approach is utilized for the detection process. The experimental results show the proposed SELUDAE attains better accuracy of 98.43% and 99.31% on NSLKDD and UNSW-NB15 dataset when compared to the existing methods like Deep Convolution Neural Network (DCNN) and Bidirectional Independent Recurrent Neural Network (BiIndRNN).