Abstract Distributed Denial-of-Service (DDoS) attacks remain one of the most critical cybersecurity threats, causing severe disruptions to network services, financial losses, and compromised system availability. With the increasing complexity and volume of network traffic, traditional detection methods often struggle to accurately identify malicious activities in real time. Deep learning techniques have demonstrated significant potential in enhancing DDoS attack detection due to their ability to learn complex traffic patterns and automatically extract relevant features from large-scale datasets. However, the high dimensionality of network traffic data can lead to increased computational cost, overfitting, and reduced detection efficiency. This study focuses on optimizing feature selection in deep learning models to improve the performance of DDoS attack detection systems. The proposed approach integrates advanced feature selection techniques with deep learning architectures to identify the most informative and discriminative network traffic features while eliminating redundant and irrelevant attributes. By reducing data dimensionality, the model achieves improved classification accuracy, faster training time, and enhanced scalability for real-time intrusion detection environments. Experimental evaluation is conducted using benchmark cybersecurity datasets to compare the proposed framework with conventional machine learning and deep learning approaches. The results demonstrate that optimized feature selection significantly enhances detection accuracy, precision, recall, and F1-score while reducing computational overhead. Furthermore, the study highlights the importance of selecting optimal features in building efficient and robust cybersecurity solutions capable of adapting to evolving attack patterns. The findings contribute to the development of intelligent and lightweight DDoS detection systems suitable for deployment in modern cloud, IoT, and high-speed network infrastructures. Overall, this research provides a scalable and effective framework for improving network security through optimized deep learning-based intrusion detection mechanisms.
HOTA et al. (Sat,) studied this question.