Machine learning algorithms in context of intrusion detection

Design of efficient, accurate, and low complexity intrusion detection system is a challenging task. Intrusion detection method is a core of intrusion detection system and it can be either signature based or anomaly based. Although, signature based has high detection rate but it cannot detect novel attacks. Asymmetrically, anomaly based detection method can detect novel attacks but it has high false positive rate. Many machine learning techniques have been developed to cope with this problem. These machine learning algorithms develop a detection model in a training phase. This paper compares different supervised algorithms for the anomaly-based detection technique. The algorithms have been applied on the KDD99 dataset, which is the benchmark dataset used for anomaly-based detection technique. The result shows that not a single algorithm has a high detection rate for each class of KDD99 dataset. The performance measures used in this comparison are true positive rate, false positive rate, and precision.