Unsupervised Feature Learning for Whatsapp network Data packets using Autoencoder

Abstract

Nowadays the network traffic analyses plays a important role in network management. The network management includes Quality of Service, blocking a particular service or application with in the organization network. There are two versions of network traffic analysis existing one is encrypted and other is unencrypted. Instant Message applications such as whatsapp,viber,telegram are generating encrypted network traffic. This type of traffic can be analyzed by analyzing the behavior of network packets flow. The objectives of doing such encrypted traffic analysis include Traffic Clustering, Application Type and Protocol Classification, Anomaly Detection or File Identification. This research is focused on capturing the whatsapp data packets at router level and clustering the packets. Since the packets are captured at router level they dont have any label. It is proposed to apply unsupervised Machine Learning, Deep Learning algorithm such as K Means, PCA, Autoencoder are applied for clustering the network data packets. PCA, Autoencoder are both unsupervised learning approach for dimensionality reduction. The Autoencoder along with K Means algorithm gives good results in clustering the network packets according to their file type(jpg,pdf,mp4).