A hybrid model for detecting intrusions using stacked autoencoders and extreme gradient boosting

In the contemporary digital landscape dominated by the internet, a wide array of attacks occurs daily, driven by a large and diverse user base. The field of identifying these cyberattacks is rapidly growing and is mainly accomplished through the utilization of intrusion detection systems (IDS). The IDS is designed to continuously observe data flow and identify any potentially harmful or suspicious acts that could signal a cyberattack. Traditional machine learning (ML) techniques encounter challenges ineffectively detecting unknown attacks and dealing with imbalanced data distributions, resulting in reduced detection performance. This paper presents a hybrid IDS model that integrates an ML classifier like XGBoost with a stacked sparse autoencoder (SSAE). The low-dimensional features obtained from the SSAE are utilized for training the classifier. The experimental outcomes indicate that the model surpasses the formerly recommended approaches regarding intrusion detection and decreases the ML classifier's training and testing times. We have also evaluated our model's performance by comparing it with other advanced techniques documented in the existing literature.