Machine learning vs. conventional regression models in public hospital waste forecasting

This study aims to predict healthcare waste generation and identify associated features. It utilises monthly data from 2019 to 2023 from a representative public hospital in Bahrain. Two models were developed using the conventional multiple linear regression (MLR) and the machine learning (ML) methods. The ensemble voting regressor, which comprises a light gradient boosting machine and three random forest algorithms, emerged as the best model, demonstrating robustness and successfully explaining 90.4% of the total variance with the lowest error. In contrast, the MLR model explains 47.7% of the total variance. The key features for predicting healthcare waste generation, ranked by the ML model from most important to least important, are the number of inpatients, population, number of surgeries, and number of outpatients. These findings provide insights into the factors influencing the rate of healthcare waste generation in public hospitals, and the effectiveness of the ML model in estimating amounts of healthcare waste.