In the realm of cyber security, particularly in the area of network security, IDS has recently attracted researchers’ attention since it has been considered one of the most powerful techniques incorporated to identify harmful activity in a network system. A functional IDS examines incoming to and outgoing from the public or private network traffic data and gives warnings once malicious data are detected. Network traffic has been quickly increasing in recent years, and unfortunately conventional IDS-based solutions, which rely mostly on signatures and restrictions, are incapable to handle massive volumes of data and prohibit novel attack events and plans. Anomaly detection-based Deep Learning (DL) techniques are more adaptable and powerful with big data than conventional approaches, and this renders them appealing to scholars. The aforementioned challenges motivate us to present our technique. The major aim of this article is to implement a lightweight IDS-based Neural Network (NN) incorporating feature selection suitable for applications that do not require very high accuracy. In our experiment, we first train our model utilizing the newer dataset with higher-dimensional, CSE-CICIDS2018, to evaluate the model performance. Then, we retrain the dataset with feature selection approach, namely Recursive features elimination (Rfe), to reduce the model complexity. Finally, our research shows that our proposal outperforms the state-of-the-art in terms of model complexity and accuracy when feature selection method is applied