Confidence-Aware Trustworthy AI System for Reliable Fraud Detection Developed an advanced fraud detection system using Random Forest classification integrated with confidence estimation, entropy-based uncertainty measurement, and trust scoring to improve reliability in automated financial decision-making. The system analyzes transaction data, predicts fraudulent activities, and evaluates whether each prediction is reliable enough for automatic action or should be flagged for human review. Achieved 99.87% accuracy, 99.86% average confidence, and 0.989 trust score on 5,000 transaction samples, demonstrating a practical approach toward trustworthy AI in high-stakes financial applications. The project enhances traditional machine learning by adding a reliability decision layer, making fraud detection safer, transparent, and more suitable for real-world deployment.