Feature Extraction using LSTM Autoencoder in Network Intrusion Detection System

Abstract

IDS (intrusion detection systems) use analysis of network traffic patterns to detect incidents of hacking. It is essential to do feature extraction in order to minimize the computational cost associated with processing raw data in the IDS. Feature extraction decreases the number of features, which decreases the time it takes to train and increases accuracy. This research employs a simple LS TM autoencoder and a Random Forest to recognize intrusion attempts by IDSs. By activating and disabling various characteristics, the extent to which this feature extraction function can enhance accuracy is examined. To find out if detection algorithms are effective after feature extraction, the NS L-KDD dataset has been employed. Autoencoder hyperparameters contain the two activation functions. The loss and activation functions of the ReLU and the SoftMax have the greatest accuracy rating of any function. The use of a Long Short-Term Memory Autoencoder (LSTMAE) and a Random Forest (RF) for identifying the best features is a goal of this study. According to preliminary experimental data, classifiers that employ these variables have a prediction rate of 94.74 percent.