DiagNCF: Diagnosis Neural Collaborative Filtering for Accurate Medical Recommendation

Qingyi Pan, Jingyi Zhang

Abstract: Performing accurate medical diagnosis recommendations is crucial but also imposes new challenges. Although end-to-end deep learning methods can enhance performance, they are seldom applied in medical recommendations due to their lack of interpretability. On the other hand, traditional statistical methods are easily interpretable but often limit in performance. In this study, we propose a novel framework called Diagnosis Neural Collaborative Filtering (DiagNCF) to improve performance. DiagNCF leverages the advantages of both matrix factorization and deep models. It decomposes the final prediction into a linear matrix factorization component, concentrating on local linear correlations, and a non-linear correlation that captures the implicit relationships between diseases and laboratory test results. We conducted experimental results on several medical datasets, specifically the MIMIC3 dataset, demonstrate that DiagNCF effectively provides accurate and efficient recommendations.

External IDs:dblp:conf/icic/PanZ24