A novel dimension reduction based method to detect network intrusions

There is a steep rise in the trend of the utility of internet technology nowadays. This tremendous increase ushered in a massive amount of data generated and personal information has been stored in the cloud. For apparent reasons, undivided attention could cause huge economic losses and cause serious security issues. Therefore, effective security solutions for detecting and combating cyber-attacks have become increasingly important. In addition, intrusion detection systems can be divided into several detection methods based on data type and analysis methods. However, no one method is absolutely effective in a variety of environments. For different types of cyber-attacks, various intrusion detection systems have been developed using artificial intelligence or machine learning techniques, but with the change in attack patterns, the intrusion detection performance of these systems needs to be greatly improved. In this study, the anomaly detection concept is used to propose a new local kernel-principal component analysis (LKPCA) method to improve classification performance. In addition, we are developing two multi-class learning strategies for binary classifiers to build multi-class classifiers. The analysis was duly conducted on the benchmark NSL-KDD and KDD datasets. The proposed method is validated using cross-validation and compared with existing feature extraction methods. This method outperformed over the existing intrusion detection methods used for the dataset.