Abstract—In the paper, Random Forest algorithm (RF), bagging and error-correction output code model (ECOC) were employed to predict the clinic expenditure of infantile patients with data consisting of records extracted from a hospital system. Throughout the modelling, the training set utilized 80% of the records selected from the original data set in random and the rest of data were used in the test set. The RF received superior predictive accuracy than bagging and ECOC, with RMSE being 0.138, R² being 0.928, |R| being 0.885, and Acc ± 1 being 85.5%. Additionally, both RF and bagging obtained impressive performances on different types of charges, achieving over 80% accuracy on average. Besides, among different types of information, clinic features obtained better results, with RMSE being around 0.2, R² being more than 0.8, |R| being larger than 0.7 and Acc being nearly 65%. In comparison, the random forest and bagging performed slightly better than ECOC models in most fields. To summarize, all three types of methods could obtain good performance during prediction, with accuracy of nearly 80%, and clinic features could provide models with higher accuracy among all fields of information.

Index Terms—hospitalization costs, forecasts, random forest, error-correction output code, multiple classification