Intrusion detection systems (IDS) are pivotal for network security, necessitating high accuracy and low false alarm rates. This paper focuses on optimizing IDS performance through Recursive Feature Elimination for feature selection and employs specialized machine learning algorithm, i.e. quantized deep neural network (QDNN-IDS), on the RT-IoT2022, CICIDS2017, CICIDS2018, and OD-IDS2022 datasets, to focus on Denial of Service (DoS) and Distributed-DDoS attacks for model training. The paper compares the accuracy and precision-based performance metrics for other models such as LSTM, CNN-LSTM, F-16, and F-8. Through an extensive evaluation on state-of-the-art datasets, the paper demonstrates the efficacy of machine learning algorithms for intrusion detection. Standalone models along with F-8 are rigorously assessed, offering insights into their effectiveness and highlighting the potential of F-8 to improve IDS accuracy. This research work contributes to understanding the strengths and limitations of different machine learning approaches and underscores the benefits of F-8 in enhancing intrusion detection systems’ accuracy.