A Cyber Intrusion Detection Method based on Focal Loss Neural Network
Zhonghao Cheng, Senchun Chai · 2020
In recent years, the applications of cyber security have been more and more widespread. Intrusion detection system as the main research direction of cyber security has attracted much attention from industrial and academic area. The performance of traditional prior knowledge based methods is degraded significantly when the system is placed in the great variable environments. Machine Learning detection method, which depends on the neural network, has high flexibility in complex environments. The sample imbalance problem of intrusion detection dataset usually confuses the engineers and researchers. In this paper, we propose an intrusion detection method which is focal loss based neural network to reduce the influence of sample imbalance problem. The focal loss pays more attention on the wrong predicted samples. In other words it shrinks the affects of well trained large sample categories in the total loss. In order to illustrate the performance of the proposed method, we implement two intrusion detection systems while training in different loss functions: cross entropy loss and focal loss. The experiment results show that the proposed method can effectively increase the detection performance of few sample categories.