Creating A Patient Data Redundancy Detection Model using Deep Learning Methods

Redundancy of patient data in healthcare facilities often arises because of double entry. Although Standard Operating Procedures (SOPs) have been enforced, patient conditions and situations during input tend to dominate input results. Entity matching is a potential solution for overcoming this problem. However, with the growth in data volume and the complexity of relationships between entities, entity matching often faces constraints related to the quality of results, time efficiency, and resources. As a solution, deep learning offers an alternative approach to entity matching, especially in situations where the data have a large size and a wide variety of entity representations, such as patient data. This study aimed to improve the validity of patient data by applying deep learning to entity matching. Four methods were planned for testing: SIF, RNN, Attention, and Hybrid. Although previous research has shown that the hybrid method significantly outperforms the other three methods in contexts with similar datasets, our findings demonstrate that the RNN method is more precise. From a dataset of 17,075 entries, we successfully identified 199 matching data pairs (positive labels), highlighting a significant initial imbalance. We employed an undersampling method to manage this imbalance. The results were promising, with an F1-score of 98.87.