Abstract

Distributed Denial of Service (DDoS) attacks continue to be a serious cybersecurity risk, interfering with services and resulting in large losses. Advanced solutions are required since traditional detection techniques frequently fall short in addressing the increasing complexity of these threats. This study proposes a deep learning-based approach for efficiently detecting DDoS attacks using Long Short-Term Memory (LSTM) networks. In order to address class imbalance, standardize features, and organize the data for sequential analysis, we preprocess the CIC-DDoS2019 dataset, which comprises 191,694 network flow records with 88 attributes. Using two LSTM layers with dropout regularization to prevent overfitting, our LSTM model achieved high performance on the CIC-DDoS2019 dataset (AUC-ROC=0.9999, testing accuracy=99.94%), demonstrating strong dataset-specific separability. With few false positives (0.46%) and false negatives (0.008%), the model demonstrates strong generalization in real-world scenarios including enterprise, cloud, and campus networks. Real-time processing capability (8ms per sample), strong detection confidence (99.7% of attack samples > 0.999), and temporal pattern recognition are among its main advantages. Comparative analysis demonstrates better performance than conventional techniques such as machine learning classifiers (95.1% accuracy) and signature-based detection (85.2% accuracy). The study demonstrates how well LSTM networks separate malicious from benign traffic, providing a scalable remedy for contemporary DDoS mitigation.