A new layer-aware graph attention and patch transformer framework improves multi-class attack classification in Industrial IoT. The system combines graph neural networks with transformers to better identify complex network intrusions. This architecture helps security practitioners reduce false positives in critical infrastructure monitoring.