A Comparison of the Daily Income of a Hospital by using Machine Learning Methods and the ARIMA Model.

Abstract

This research compares the daily income of a hospital during an epidemic situation using Machine Learning and ARIMA Models. Tableau Desktop is used to create data reports and R studio is used to analyze data. By studying the daily data of a large private hospital in western Bangkok, the data is divided into 2 periods: 1. The normal situation during the year 2016 to 2019, combined with the year of the coronavirus outbreak the pandemic year 2021, and 2. The irregular situation of the coronavirus 2019 outbreak pandemic from 2020 to 2021. Using 6 machine learning model methods: Simple Linear Regression Model, Multiple Linear Regression Model, Polynomial Regression Model, Support Vector Regression Model, Decision Tree Regression Model, and Random Forest Regression model, both periods of the Machine Learning. Divide the training data and the test data in an 80:20 ratio. Compared with the ARIMA model. Where the ARIMA model during normal situations. Divide the training data is the years 2016 to 2019, and the test data is the year 2021 and the irregular situation period. coronavirus epidemic. Divide the training data is the years 2020, and the test data is the year 2021, the statistics used to measure the model's performance and accuracy. Considering the Mean Absolute Percentage Error: MAPE, and Mean Squares Error: MSE, by the models that gave the lowest MAPE and MSE values. The results of the research showed that in Machine Learning with independent variables involved there is a lower error in the forecast model than in the ARIMA model with no independent variables involved, both during normal situations and during irregular situations in coronavirus epidemic. The presence of independent variables involved significantly affects the reduction of the mistakes of the model used in the forecast and can group the days of the day as follows: Monday and Wednesday the model with the suitable forecast performance is the Multiple Linear Regression model in irregular situations of the coronavirus 2019 outbreak pandemic, giving the value of the lowest MAPE(MSE) as 0.00161% (3.60E-09MB²) and 8.13E-15% (1.10E-31MB²), respectively. Tuesday Thursday and Friday the model with the suitable forecast performance is the Random Forest Regression model in the normal situation giving the value of the lowest MAPE(MSE) as 5.04484% (0.04521MB²), 3.17105% (0.00979MB²) and 4.78181% (0.02949MB²), respectively. Saturday and Sunday the model with the suitable forecast performance is the Random Forest Regression model in irregular situations of the coronavirus 2019 outbreak pandemic, giving the value of the lowest MAPE(MSE) as 6.76110% (0.05932MB²) and 4.51761% (0.02101MB²), respectively. When there is another epidemic crisis in the future. The daily appropriate forecasting model derived from this research can be used as the default model for forecasting hospital daily revenues and enabling the strategy to continue even in critical situations.