Advancing democratic processes in Ecuador: A case study on neural network‐driven OCR for election report verification

Background In Ecuador, scepticism surrounding electoral outcomes underscores the need for a reliable system to ensure transparent election results. Manual verification demands a more efficient approach due to the vast volume of election reports. This research introduces an automated system leveraging Artificial Intelligence to process results from Ecuador's three recent national elections. Methods The system, designed with a three‐layer architecture, extracts, processes, analyses, classifies and compares election results. We thoroughly analysed the National Electoral Council of Ecuador (CNE) web pages for effective data extraction and processing. Rigorous unit and acceptance tests validated the system's functionality. A classifier model, trained using data augmentation techniques, achieved a 98% accuracy rate. Results While the system boasts high efficiency, we identified three errors, accounting for less than 5% of the total fields processed. Notably, the quality of scanned reports and illegible handwritten numbers posed challenges for the classifier. Conclusions The system's deployment by an authorized entity in Ecuador could enhance the CNE's information verification. Despite some errors, the system's potential is clear. Future work includes refining classifiers, verifying officer signatures, and expanding the system's scope, aiming for a more transparent electoral process.