An Effective Hyperparameter Selection for Deep Learning Algorithm in Intrusion Detection System

Abstract

With the great success of artificial intelligence, machine learning and deep learning have been applied to intrusion detection as the core methods in recent years. The primary focus of many researches is on how to improve the performance by tuning the deep learning model and the associated hyperparameters. Hyperparameters are critical factors for deep learning-based method which includes how many layers should be used in the model; how many neurons are there for each layer; and how to set up the learning rate properly. However, most of studies set the hyperparameters depending on personal experience or using inefficient brute force exhaustive search. In this paper, we will present a self-adaptive method to decide the setting of hyperparameters that integrated deep learning and differential evolution. The proposed scheme is then be used for intrusion detection system. To evaluate the effectiveness of the proposed method, the performance of the scheme is compared to those of other machine learning and deep learning methods. Three performance metrics in terms of accuracy, recall and precision are used. Results show that the proposed algorithm can find the best hyperparameters setting for deep learning model than other methods compared in this paper.