Data Interpolation for Deep Learning based Encrypted Data Packet Classification

In this paper, we propose to apply deep learning for encrypted data packet classification. Due to the complexity of deep learning, it is inefficient to process the entire data packet, which is usually padded with many zeros. In order to reduce data size for classification, while maintaining high accuracy, we develop data interpolation schemes to process data packets with various lengths to a fixed size. In particular, three data interpolation schemes based on nearest value, bilinear and bicubic interpolation are proposed. Experiments are conducted using an open source dataset. The evaluation results demonstrate that our data interpolation schemes can be applied to process input data for higher computational efficiency without losing classification accuracy.