Signature Based Network Intrusion Detection with Deep Autoencoder

Abstract

Electronic mail, a quick, easily accessible and extremely affordable communication device, has become one of the key ways of communication which enables communication in business life due to these features. This is also mirrored in the growth in the use of electronic mail, theft and spam mails received outside the person's advertisement order. Spam mails are open to workers every day as the amount of mail and often even more than regular mail. These mails are tracked , monitored and collected causing a significant loss of energy, manpower and financial capital, which tends to adversely affect the life of industry. For these purposes, studies aimed at automatically detecting spam e-mails are up to date and relevant. For this research, an autoencoder based deep learning system is proposed for the detection of spam mail, as is one of the current machine learning approaches. Autoencoders are a multilayer representation of the same data collection for inputs and outputs. Autoencoders consisting generally of three layers, using less hidden input and output layer neurons, reduce the size of the data set by nonlinearly mapping the input data set to the output data set. The data obtained from an auto encoder's hidden layer acting as nonlinear PCA are input data reduced to a smaller space. By connecting several autoencoders like a stack, the deep learning architecture is built. The deep learning encoder is obtained by connecting the training layer with class information on the last line. The classificator using a deep learning auto-encoder has shown an incredibly high performance rate of 98 percent.