Beta-hCG test demand forecasting using stacking ensemble-learning and machine learning approaches

Abstract

Demand forecasting is essential for decision-making, since these forecasts are important inputs for strategic management decisions. In this context, the contribution of this study is to propose a hybrid forecasting framework that combines machine learning (ML) models and a stacking ensemble-learning (STACK) approach to forecast the Beta-hCG test demand using a multi-day ahead forecasting strategy. The experiment consisted in comparing the performance of the STACK strategy with the ML models using statistical performance measures. The results show that the STACK model was the most accurate forecaster for 1, 30, and 45 days ahead, while the Generalized Linear Model was the most accurate for 15 days, and Gaussian Process Regression for 60 days. In summary, the STACK model outperformed the compared models in the analyzed forecasting horizons with an average of improvement performance index ranging from 0.69% and 28.38%. Indeed, the proposed forecasting framework provides forecasts that support decision-making in diverse strategic departments in the company, such as sales, marketing, manufacturing, and logistics departments.