Injury narrative text classification using factorization model.

Area of Science:

• Medical Informatics
• Natural Language Processing
• Machine Learning

Background:

• Emergency department (ED) data collections contain valuable narrative text detailing injury circumstances.
• Manual classification of these narratives is time-consuming and labor-intensive.
• Automated classification using machine learning (ML) offers a solution to streamline this process.

Purpose of the Study:

• To explore advanced machine learning techniques for automated classification of injury circumstances from ED narrative data.
• To compare the performance of Matrix Factorization approaches against existing methods like Naive Bayes.
• To investigate the impact of parameter settings on classification accuracy for medical text data.

Main Methods:

• Implementation of Matrix Factorization techniques, specifically Non-Negative Matrix Factorization (NMF).
• Inclusion of a learning enhancement process to improve classification performance.
• Comparative analysis with other classification approaches using a medical text dataset.

Main Results:

• The proposed Matrix Factorization approach, particularly NMF, demonstrates superior performance compared to existing methods.
• With optimal parameter selection (dimension k), the NMF model achieved a 10-fold cross-validation (CV) accuracy of 0.93.
• Analysis highlights the sensitivity of classification results to parameter settings in medical text classification.

Conclusions:

• Matrix Factorization, enhanced with a learning process, is a highly effective method for classifying injury circumstances from ED narratives.
• This approach significantly reduces the need for manual data classification, offering efficiency gains.
• The study underscores the importance of parameter tuning for maximizing the performance of ML models in medical text analysis.