This research is a comparative analysis of the application of different machine-learning methods to health care data to predict home care usage in consultation with patient partner involvement. The data used are from the interRAI Home Care assessment after instrument, collected in central British Columbia, Canada. The original data set used contains 837,536 records, gathered from 2010 to 2019, and 423 attributes. The model is developed for predicting the average hours per day usage of home care services in the three weeks following an assessment using different regression and classification methods. For regression, I used multiple linear model, lasso, ridge, decision tree, and ensemble methods, where the last appeared as the most promising. For classification, I used KNN, logistic regression, decision tree, and ensemble methods. Apart from the machine learning algorithms, both patient partners and health care experts participated and provided feedback regarding home care practices and issues. These formed essential elements in designing the research questions, selecting variables, and improving the models. The ensemble methods, namely Random Forests and Bagged trees, are found promising for both regression and classification problems. The Random Forests has achieved the largest R2(0.53) in predicting the average hours per day. For classification, the largest accuracy and ROC AUC scores are 0.96 and 0.97 respectively, obtained from the Random Forest and Bagging algorithms.
