AI System for Automated Insurance Claim Validation

This thesis focuses on automating the task of extracting data from vehicle images in the insurance claims process. It proposes the design, development, and evaluation of an AI system for a major Czech insurance company. The system uses a fine-tuned YOLOv8 model for object detection and a PaddleOCR model for text recognition to extract key information from claim images: license plates, odometers, and VIN. Modification and verification rules are defined to ensure the validity of the extracted texts. The developed system undergoes validation to assess its overall performance. The proposed system achieved an overall accuracy of 80.4 % in extracting key vehicle data. License plate extraction proved particularly successful, reaching an accuracy of 98 %. This translates to considerable cost savings and a more streamlined customer experience, leading to higher customer satisfaction.