Hybrid Fault Classification System — XGNN Project
Classification
I engineered and annotated a synthetic fault dataset to train and evaluate an Explainable Graph Neural Network (XGNN) for hybrid fault classification and localization. I designed and labeled 2,400 simulated distribution network scenarios across 8 distinct fault classes based on electrical features. I integrated model explainability and root-cause localization to ensure physics-aligned and transparent classification. • Manually stratified scenarios for balanced train/validation/test splits
• Validated performance using explainable AI and integrated gradients
• Used conformal prediction for uncertainty-aware case routing
• Documented Shannon Entropy hybrid decision framework
I engineered and annotated a synthetic fault dataset to train and evaluate an Explainable Graph Neural Network (XGNN) for hybrid fault classification and localization. I designed and labeled 2,400 simulated distribution network scenarios across 8 distinct fault classes based on electrical features. I integrated model explainability and root-cause localization to ensure physics-aligned and transparent classification. • Manually stratified scenarios for balanced train/validation/test splits • Validated performance using explainable AI and integrated gradients • Used conformal prediction for uncertainty-aware case routing • Documented Shannon Entropy hybrid decision framework
2026 - 2026