A Fusional Intrusion Detection Method Based on the Hierarchical Filtering and Progressive Detection Model

Abstract

The imbalanced distribution of intrusion detection dataset will usually result in low detection rates for the minority categories and more unreported attacks. To reduce the impact of imbalanced dataset on classification performance, a fusional intrusion detection method based on the hierarchical filtering and progressive detection model (HFPD-IDS) is proposed. This method first uses the idea of binary classification to preliminarily filter and identify the normal and abnormal data, and then uses a convolutional neural network model for progressive dimensionality reduction detection for the initially filtered abnormal data. The learning process during the progressive detection, which is entirely based on abnormal data, avoids the interference of normal data and integrates detection method different from the initial filtering process to improve the detection rate for minority categories. Finally, the normal dataset identified in the initial filtering process will be corrected to further identify the abnormal data that has been misjudged as normal data. The experimental results show that this method can significantly improve the detection rate of minority categories data without affecting the detection performance of majority categories data.

Keywords

• Intrusion detection
• binary classification learning
• convolutional neural networks
• feature selection
• detection rate