Reducing Features Within an Extensive Network of Medical Subject Headings Metadata to Enhance Deep Predictions

Abstract

Feature reduction in a large amount of Medical Subject Headings (MeSH) metadata for deep prediction involves selecting a subset of relevant features to improve the efficiency and effectiveness of deep learning models. MeSH is a controlled vocabulary thesaurus used for indexing articles in the life sciences and biomedical fields. We analyze a disease-symptom network exploiting MeSH metadata and configure a deep model for disease prediction based on symptoms. Dimension reduction techniques have yielded positive results in optimizing a large amount of Medical Subject Headings (MeSH) metadata for deep prediction with good accuracy. Therefore, our result highlights that decrease in the severity or degree of symptoms associated with a disease correlates with an improvement in the accuracy of disease prediction. This finding may have important implications for disease prediction models in healthcare: interpretation of results, clinical significance, practical implications.