The Study of Integrated Deep Learning Network Anomaly Detection Mechanisms

• Main Authors: TSUNG-EN WU, 吳宗恩
• Other Authors: WEN-SHENG JUANG
• Format: Others
• Language: zh-TW
• Published: 2019
• Online Access: http://ndltd.ncl.edu.tw/handle/bh339m

碩士 === 國立高雄科技大學 === 資訊管理系 === 107 === In this thesis, we proposed an integrated deep learning method for analyzing network abnormal behavior or distinguishing the attack mode through the network logs. This research uses the conn.log provided from Bro-IDS for analyzing the behavior. We use the deep learning mechanisms without the need of the rules from persons for automatic learning the features of network abnormal behavior. This can help the information staff for doing timely response for intrusion prevention. We use various deep learning architectures in our proposed method. Multi-layer perceptron (MLP) can handle nonlinear data and is the most widely used architecture. Convolution neural network (CNN) in the spatial analysis is more outstanding. Long short-term memory (LSTM) is good for the processing of time series and only needs some information features. By our proposed ensemble learning, we can completely and accurately analyze and detect various types of network attacks in the cyber world.