Evaluating resampling methods and structured features to improve fall incident report identification by the severity...

Area of Science:

• Medical Informatics
• Machine Learning
• Data Science

Background:

• In-hospital falls are a significant safety concern, and accurately classifying their severity is crucial for targeted interventions.
• Existing methods often struggle with imbalanced datasets and underutilize structured information within incident reports.

Purpose of the Study:

• To develop and evaluate an incident report classification (IRC) framework for improved classification of fall incident severity levels.
• To address data imbalance issues and incorporate structured features to enhance machine learning model performance.

Main Methods:

• An IRC framework was developed, incorporating text preprocessing, feature extraction (bag-of-words, structured text, and clinical features), and resampling techniques.
• Machine learning models were trained and validated using a stratified cross-validation approach.
• System performance was evaluated using F1-measure, precision, and recall.

Main Results:

• The proposed IRC framework, considering data imbalance and structured features, achieved a mean macro-averaged F1-measure of 0.733, outperforming other settings.
• Incorporating structured features and resampling significantly improved the F1-measure for rare classes by 30.88% and the overall macro-averaged F1-measure by 8.26%.
• The random forest algorithm combined with random oversampling demonstrated superior performance.

Conclusions:

• Structured features are vital for accurate fall incident severity classification.
• Resampling methods effectively rebalance class distribution, enhancing machine learning model performance.
• The IRC framework automates the identification of fall incident reports by severity level, improving safety management.