A Keyword-Enhanced Approach to Handle Class Imbalance in Clinical Text Classification.

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Natural Language Processing

Background:

• Deep learning models show promise for healthcare text classification.
• Imbalanced datasets, common in healthcare, hinder model reliability and performance on rare classes.
• Robust performance on rare classes requires additional model constraints when training data is limited.

Purpose of the Study:

• To present a novel strategy for improving deep learning model accuracy on rare classes within imbalanced healthcare datasets.
• To introduce a method for incorporating keywords as supplementary training data.
• To enhance the reliability of text classification models in real-world healthcare applications.

Main Methods:

• Assembled a set of keywords, including short phrases, associated with each class.
• Integrated keywords as additional data during each batch of model training.
• Developed a training loss function with contributions from both raw data and keywords.

Main Results:

• Demonstrated a substantial increase in model performance for rare classes in cancer pathology report classification.
• Showcased the effectiveness of keyword incorporation in boosting accuracy for underrepresented categories.
• Analyzed the impact of keywords on model output probabilities for bigrams.

Conclusions:

• Keyword incorporation is an effective strategy to address data imbalance issues in healthcare text classification.
• The proposed method offers a straightforward approach to identify and mitigate model difficulties related to limited training data.
• This technique enhances the robustness and reliability of deep learning models for critical healthcare applications.