Comparison of Prophet with Machine Learning Regression Models for Long and Short-Term Manufacturing Forecasting

This paper compares Facebook's Prophet with classic regression-based machine learning algorithms for short-term and long-term forecasting in a client's manufacturing system. The study uses advanced forecasting methods on four different real-world datasets with rich temporal features and complex patterns to provide insights for businesses in the manufacturing industry. The results showed that Prophet outperformed other models for short-term forecasting with an error percentage of 36.82% when the data trends changed rapidly over a shorter period. On the other hand, tree-based algorithms performed better for short-term forecasting with more stable trends in the training data. For long-term forecasting, the Least Absolute Selection and Shrink-age Operator regressor and Random Forest regressor emerged as the superior models, with error percentages of 39.60% and 39.60% respectively. The computational analysis demonstrates that Bayesian Ridge and Least Absolute Selection and Shrinkage Operator regressors perform better than other models. This study provides guidance to practitioners and decision-makers in the manufacturing industry on selecting applicable forecasting models based on the relevant time horizon.