Classification Of Encrypted Network Traffic

Abstract

Now-a-days internet is everywhere and
responsible for generating different types of network
traffic such as Email traffic, video streaming, browsing
and many more. Analysing or predicting different types
of traffic can help in knowing which type of traffic is in
more usage and which is in less usage. There are many
deep and machine learning algorithms are available to
classify different network types but they lack support of
Network in Network model and Global Average Pooling
(helps in generating one feature map for each
corresponding category of the classification task and
this features mapping help in reducing model
parameters and better classification accuracy). Deep
Network in Network model will use parallel different
layers to train Network Packet Header and Packet body
whereas existing algorithms were using same layers to
train both header and body. This paper presents a
neural network model with deep and parallel networkin-
network (NIN) structures for classifying encrypted
network traffic. Comparing with standard
convolutional neural networks (CNN), NIN adopts a
micro network after each convolution layer to enhance
local modelling. Besides, NIN utilizes a global average
pooling instead of traditional fully connected layers
before final classification, which reduces the number of
model parameters significantly. In this proposed
method, deep NIN models with multiple MLP
convolutional layers are built to map fixed-length
packet vectors towards application or traffic labels.