Automating Clinical Document Classification: AI Solutions for Enhanced Healthcare Decision Support

Labeling documents using traditional methods requires a lot of manual work. To classify clinical documents specifically, hospitals must assign expensive medical staff to manually classify documents, which is prone to errors and delays. To solve this problem, machine learning methods have proven their capability to perform the task automatically. Our research aims to fill the gap in the literature by exploring machine learning models to handle narrative document classification. The methodology consists of steps organized within a defined pipeline to handle the data. The input data is clinical documents and the output is the label class for this document. A comparative analysis of the most commonly used classification models and their feature engineering techniques was conducted including traditional algorithms, sequence models, and transfer learning using BERT transformers. The outperforming model (accuracy 0.89, fl-score 0.886) proves that integrating such tools enhances healthcare providers' performance.