Abstract

The major problem computer network users face with data that is in storage, in transit and being processed is unauthorized access. This unauthorized access normally results in loss of confidentiality, integrity and availability of such data which requires an accurate Intrusion Detection System (IDS) is put in place for every information system. Machine learning and deep learning model such as autoencoder have been developed by many researchers to improve existing IDS. However, the issue of accuracy of this model still remains a major research issue. The aim of this study, therefore, is to propose Correlation-based Feature Selection and Autoencoder (CFS-AE) to improve detection accuracy and reduce the false alarms associated with current anomaly IDS. The feature selection of the NSL-KDD dataset used to train and test our model is first carried out. Autoencoder is then used to classify the data traffic into attack and normal. The results from our experimental study recorded an accuracy of 94.32% which is better compared with existing and previous systems.