A Machine Learning Approach for Predicting Patient's Length of Hospital Stay with Random Forest Regression

Predicting the length of stay (LOS) of patients can be incredibly useful for an efficient hospital resource management. Numerous hospitals worldwide are confronted with limited resources to accommodate hospitalized patients. By assuring optimized resource utilization, LOS forecasting can benefit all the stakeholders with improved treatment planning and expenditure estimation. However, traditional hospital management systems fail to estimate the duration of patients' stay at an early stage, resulting in various negative consequences. Therefore, this study proposes a machine learning approach for analyzing patient data and building a reliable prediction model incorporating Random Forest Regression Model to estimate the length of a patient's stay in the hospital. The data utilized in this study is a hospital discharge dataset containing records with various types of patient information. Two types of feature selection methods (PCA and Chi-square) and interquartile range based outlier elimination strategy have also been employed in the model for efficient prediction. To validate the performance of the proposed method including the impact of feature prioritization and outlier elimination, the dataset has been applied to ten different regression models as well as deep learning techniques after requisite data pre-processing. Furthermore, a comparison of multiple prediction models was conducted using various performance metrics in a variety of cases to determine the best performing regression model; in which the Random Forest Regression model outoerformed other models.