INTEGRATING GNN-LSTM IN SOFTWARE DEVELOPMENT FOR ENHANCED SECURITY AND PERFORMANCE OPTIMIZATION

Abstract

In modern software systems, managing both security and performance has become increasingly complex,
especially in distributed environments. Traditional methods often fail to address the dynamic nature of these
systems, relying on static rules that do not adapt well to evolving behaviours, resulting in delayed detection and
response to security and performance issues. To overcome these limitations, a new approach integrating Graph
Neural Networks (GNN) and Long Short-Term Memory (LSTM) networks is proposed. The GNN component
captures the spatial relationships between system components, while the LSTM model predicts temporal patterns,
enabling more accurate and real-time detection of anomalies and security breaches. The results demonstrate that
this approach reduces undetected risks by 50% and significantly improves alert time-to-detection compared to
traditional methods. However, scalability remains a challenge, as detection times increase with larger graph sizes.
These findings suggest that while the integrated GNN-LSTM framework offers considerable improvements in
security and performance, further optimizations are necessary to handle larger-scale systems effectively. This
approach offers a promising direction for enhancing both the security and efficiency of modern software systems.