Autoencoder-based Deep Learning Approach for Intrusion Detection System using Firefly Optimization Algorithms

Intrusion Detection System (IDS) is software that specializes in analysing activities that has taken place on a computer system and aims to uncover indications on whether the computer system has been maliciously misused. IDS are majorly classified into two categories such as anomaly-based and signature based. Signature based systems rely on having an understanding of manually identified patterns for which rules is defined for detecting them. Anomaly-based systems are systems that attempt to automatically identify anomalies from examples of normal data and perhaps abnormal data. Traditionally, several machine learning algorithms are used to find the anomalies in the network data. In last decade, researchers are shifted the attention to deep learning techniques to detect the anomalies in the different types of standard datasets. Unlike classification and regression tasks, there is no straightforward way to program neural networks to detect anomalies. In this work, a deep learning technique of autoencoders is used to detect the anomalies in the CSE-CIC IDS dataset. The autoencoders contains encoder and decoder networks. The encoder network compresses the input samples and decoder network reconstruct input from the compressed input samples. The autoencoder identify the anomalies based on the reconstruction error which is the amount of difference in original input sample and reconstructed sample. When the reconstruction error is higher than the threshold value the connection is treated as anomaly. The performance of autoencoder mainly depends on the features that are used in the encryption process of input samples. In this work, a firefly optimization algorithm is used as a feature selection algorithm to find the most relevant features for the experiment. The proposed autoencoder based deep learning approach attained best accuracy of 99.2% for intrusion detection.