Feature selection and classification in supporting report-based self-management for people with chronic pain.
Yan Huang1, Huiru Zheng, Chris Nugent
1Computer Science Research Institute, School of Computing and Mathematics, University of Ulster, Jordanstown, UK.
Summary
Machine learning simplifies chronic pain management by identifying key self-report questions. This approach optimizes treatment evaluation and supports patient self-management effectively.
Area of Science:
- Biomedical Informatics
- Pain Management
- Machine Learning Applications
Background:
- Chronic pain significantly impacts physical and emotional well-being.
- The integrated biopsychosocial approach is standard for chronic pain management.
- Self-report questionnaires are crucial for evaluating treatment outcomes but can be lengthy.
Purpose of the Study:
- To apply machine learning for analyzing chronic pain self-report data.
- To identify an optimal subset of questions for supporting self-management.
- To develop a classification model for differentiating chronic pain treatment stages.
Main Methods:
- Utilized machine learning to analyze self-report data from integrated biopsychosocial treatment.
- Applied four feature selection methods to rank questionnaire items.
- Employed four supervised learning classifiers to assess classification performance with reduced question sets.
Main Results:
- No significant differences in classification accuracy or AUC were found between feature ranking methods.
- Significant differences were observed between classifiers for each ranking method (p < 0.001).
- A multilayer perceptron classifier achieved 100% accuracy and an AUC of 1 using an optimized ten-question subset.
Conclusions:
- Machine learning can effectively reduce the burden of chronic pain self-reporting.
- An optimized subset of ten questions, analyzed by a multilayer perceptron, accurately identifies treatment stages.
- This approach supports enhanced self-management and efficient treatment evaluation in chronic pain care.
