Deep Learning for Encrypted Traffic Classification and Unknown Data Detection

Irrespective of the growing usage of encryption techniques to maintain privacy during Internet communications, users of mobile devices continue to face privacy and security issues. This research proposes a unique Deep Neural Network (DNN) based on a user activity detection framework to recognize certain user actions from an encrypted Internet traffic stream that is sniffed, which are called in-app activities. Maintaining the security and privacy of data transfer requires encrypted communication. Because it keeps hackers from intercepting private data that they could access without permission, it is crucial to maintain the security of our networks. Modern communication systems are seeing an increase in encrypted network traffic, which makes network management and security more difficult. To overcome this obstacle, Due to the inability to examine the content of the packets and the lack of clear visibility into their contents, machine learning models have been used to classify encrypted communication with varying degrees of success. More productive research has started on creating machine learning models for identifying encrypted payloads without explicitly analyzing their contents to address this problem. One of the difficulties is the vast number of apps; it is nearly hard to gather and train a DNN model with all the data that might come from them. As a result, in this work, we utilize the DNN output layer’s probability distribution to filter out data from applications that weren’t considered during model training (i.e., unknown data). The suggested approach uses a time-based strategy to partition the activity’s traffic flow to identify in-app actions by analyzing a small portion of the activity’s traffic. Our experiments reveal that the DNN-based framework can recognize previously learned in-app behaviors with an accuracy of 90% or higher and can identify them with an average accuracy of 79%.