Improving the Classification Performance for Network Security Detection

• Main Authors: Chou,Chia-En, 周加恩
• Other Authors: 楊棋堡
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/82610536302212143679

碩士 === 國防大學理工學院 === 資訊工程碩士班 === 100 === In the field of information security, intrusion detection systems play an important role in preventing the illegal invasion. Recently, research on building intrusion detection system trends to apply the data mining technique, which trains the learning algorithm by learning the classification rules from data automatically. The classification model is then established to identify normal or abnormal behaviors without consuming the manpowers on analyzing the data. However, the data mining-based intrusion detection systems still face many challenges, such as the effect of data outliers, the feature selection, and the lower classification accuracy. In this thesis, the Outlier Deletion algorithm, the Multiple Feature Selection algorithm, and the Class Fully Weighted algorithm were proposed to overcome those challenges mentioned previously. The OD algorithm was used to obtain the representative training data in the data preprocessing phase. The Multiple Feature Selection algorithm was used to find out the best feature space for the classifiers by selecting features during the data attribute selection phase, and then the training models of the classifiers could be well trained. Furthermore, the ten fold-cross validation was applied to evaluate the performances of the classification models to select the classification models with best recalls for each class. The selected classification models were used as the weighted classification models in the Class Fully Weighted algorithm to predict the classes of test data. Finally, the classification performance of the intrusion detection system could be enhanced by the inference results of the Class Fully Weighted algorithm. Keywords: IDS, Data Mining, Feature Selection, Outlier , Ensemble System.