Encrypted video identification - Normalization and activation function perspective

Video identification in encrypted network traffic has become a challenge for researchers in the past few years. However, the accuracy of the identification depends on various factors. This paper highlight the impact of data normalization techniques and the activation function of the hidden and dense layer on accuracy. For this purpose, we tried different normalization techniques for data normalization. Upon selecting the best data normalization technique, we first modified the activation functions of hidden layers. After selecting the most suitable activation function for hidden layers, we modified the activation function of the dense layer. The results show that the Min-Max data normalization technique, the ReLU function on hidden layers, and the Soft-max activation function on dense layer is the most suitable combination for video identification in network traffic. The configuration achieves an accuracy of 96.38%.