Research Team Publishes Study Comparing AI Models for Cloud DDoS Detection

New York, USA, August 12th, 2026, FinanceWire

Peer-reviewed research finds that an LSTM-based approach may offer a practical balance between detection accuracy and computational efficiency

 A research team has published a peer-reviewed study comparing three deep learning models for detecting Distributed Denial-of-Service attacks in cloud computing environments.

The paper, “Performance Comparison of GRU, LSTM, and RNN Models for Detecting DDoS Attacks in Cloud Infrastructure,” was published in AIP Conference Proceedings following its presentation at the Second International Conference on Advanced Network Technologies and Computational Intelligence.

The study was authored by Sunita Gupta, Parag Ravikant Kaveri, Kishan Kesari Gupta, Prashant Awasthi, and Mahaboobsubani Shaik. It evaluates Long Short-Term Memory, Gated Recurrent Unit, and Simple Recurrent Neural Network architectures using the CICDDoS2019 benchmark dataset.

Rather than examining predictive accuracy alone, the researchers assessed how different model and feature-selection combinations affected accuracy, false-negative rates, computational requirements, and potential suitability for continuous cloud-security monitoring.

Evaluating Accuracy and Efficiency

DDoS attacks occur when large volumes of malicious traffic are directed toward a network or online service. These attacks can disrupt cloud platforms, financial services, healthcare systems, e-commerce operations, telecommunications networks, and government portals.

Security systems must therefore analyze substantial volumes of network traffic while identifying malicious activity with minimal delay. A model may achieve high experimental accuracy but remain difficult to operate continuously if it requires excessive processing capacity or infrastructure resources.

The researchers evaluated Information Gain, Backward Search, and Principal Component Analysis as methods for reducing data complexity before classification.

The highest experimental accuracy reported in the study was approximately 99.75%, achieved by a Simple RNN model using Backward Search. However, this approach also required greater computational resources.

An LSTM model using Information Gain, Backward Search, and Principal Component Analysis achieved approximately 99.16% test accuracy while using fewer computational resources. Based on the study’s comparative findings, the researchers identified this approach as offering a practical balance between predictive performance and efficiency.

The LSTM-based model also produced the lowest false-negative rate among the evaluated architectures. False negatives occur when malicious traffic is incorrectly classified as legitimate, potentially allowing an attack to continue without detection.

“Our objective was not simply to identify the model with the highest accuracy,” said Prashant Awasthi, a co-author of the study. “Enterprise security systems must also account for computational overhead, scalability, processing latency, and the ability to operate continuously. These factors are important when evaluating whether an experimental model may be suitable for further production testing.”

Potential Enterprise Applications

The findings may be relevant to cloud service providers, financial institutions, healthcare organizations, government agencies, telecommunications companies, managed security providers, and enterprise security operations centers.

The researchers note that organizations evaluating AI-based cybersecurity systems should consider multiple operational factors, including:

• Detection accuracy
• False-positive and false-negative rates
• Prediction latency
• Computing and infrastructure requirements
• Scalability across changing traffic volumes
• Integration with existing security systems

The study does not establish that the evaluated model is ready for immediate production deployment. Testing was conducted using the CICDDoS2019 benchmark dataset rather than live enterprise cloud traffic.

The authors recommend further validation using additional datasets and production-scale cloud environments to assess performance under different network conditions, attack patterns, workloads, and infrastructure configurations.

About the Research

“Performance Comparison of GRU, LSTM, and RNN Models for Detecting DDoS Attacks in Cloud Infrastructure” examines how deep learning and feature-selection methods can support the detection of malicious network traffic in cloud environments.

The study was presented at the Second International Conference on Advanced Network Technologies and Computational Intelligence and subsequently published in AIP Conference Proceedings.

Official paper: [Insert AIP publication or DOI link]

AIP Publication - https://pubs.aip.org/aip/acp/article-abstract/3410/1/020069/3391954/Performance-comparison-of-GRU-LSTM-and-RNN-models

DOI - https://doi.org/10.1063/5.0330093

Google scholar profile - https://scholar.google.com/citations?user=Dq8hLYAAAAAJ&hl=en

Contact