An Adaptive Service Classification Method Based on Stacked Hybrid Auto-Encoder

In this paper, we investigate the problem of communication services classification and propose an adaptive service classification method which is based on stacked hybrid autoencoder (SHAE). A four-layer SHAE network connecting sparse auto-encoder (SAE) and contractive auto-encoder (CAE) alternately is proposed and learned into a feature extraction model from labeled services. In order to balance the complexity and adaptability, the neuron numbers of latter two layers are adjustable for the best performance. By comparing the distance from the extracted features of unlabeled service to the feature center of every kind of services which is gotten by averaging the extracted features of marked services, the service label will be obtained. The simulation results on the ISCXTor2016 dataset demonstrate that the proposed algorithm can achieve better performance than the traditional machine learning methods.