Automatically Extract Botnet Features Using an Autoencoder to Detect Botnets
| Autor: | Chen, Bor-An, 陳柏安 |
|---|---|
| Rok vydání: | 2018 |
| Druh dokumentu: | 學位論文 ; thesis |
| Popis: | 107 Botnet is one of the major threats on the Internet for cybercrimes, such as spreading spams, DDoS attack, etc. In 2016, there was a famous cybercrime by using botnet. The hacker using botnet to control lots of IOT devices launched a DDoS attack. This event made some famous network service interrupted. In the past years, there are many researcher work on botnet detection. Early, researchers focus on signature-based botnet detection. In recent years, researchers use machine learning technique like supervised learning to detect botnet. When researchers want to using supervised learning technique to detect botnet, they need to familiar with botnet and analyze the botnet dataset so they can propose the effective feature. We propose an automate botnet feature extraction method. This method can extract features from a large feature set by using autoencoder and train a classifier. With this method, we can achieve an accuracy of up to 99.6% in different data sets. Our method not only can subtract the researcher’s efforts to find effective features, can also reduce the original feature set dimension. In addition, the training of autoencoder data does not need to be labeled, and the autoencoder training can be improved with unlabeled data. Finally, we also use the autoencoder to take advantage of the characteristics of the unlabeled data and only use the general network flow to build the model to achieve anomaly detection. |
| Databáze: | Networked Digital Library of Theses & Dissertations |
| Externí odkaz: |
|