Predicting Recreational Therapy Engagement in Veterans' Long-Term Care: A Machine Learning Approach

This poster presents the first machine learning framework for predicting recreational therapy engagement among veterans in long-term care facilities. Using data from 57 veterans at the New York State Veterans Home at Oxford, we developed predictive models addressing a critical gap in evidence-based long-term care practice. Our Random Forest algorithm achieved strong performance (F1-score: 0.860 for high participation prediction), identifying activity preference diversity and facility tenure as primary engagement predictors. Key findings reveal that veterans with diverse activity interests (>4.5 preferences) and high satisfaction scores demonstrate 100% probability of sustained participation, while new residents with limited preferences face highest disengagement risk. Clinical decision rules enable early identification of at-risk veterans during routine assessments. This research provides immediately implementable tools for recreational therapy staff to optimize intervention targeting, supporting evidence-based approaches to veteran care in institutional settings. The methodology addresses class imbalance challenges common in healthcare prediction modeling.

After attending this abstract, attendees will be able to apply machine learning methodologies to predict patient engagement patterns in their own healthcare settings, including feature selection techniques and cross-validation approaches for small healthcare datasets.

After attending this abstract, attendees will be able to identify key predictive factors for recreational therapy participation among veteran populations and design targeted intervention strategies based on activity preference diversity and facility tenure patterns.