Forecasting Raw Milk Supply in Türkiye: A Comparative Study of Machine Learning and Statistical Methods

Abstract

Forecasting the supply of raw milk is an essential task for nations to determine the price and manage the demand explicitly. It is also a challenging task due to the nature of raw milk, which is influenced by numerous factors. Recent artificial learning models are used in conjunction with traditional statistical models to predict the future from time series data. However, there is no standard procedure for selecting a forecasting model. Therefore, this study performs a comparative study of machine learning and statistical methods for forecasting raw milk supply in Türkiye. First, a comprehensive set of features influencing raw milk supply is identified. Subsequently, a feature selection process is performed to eliminate the features that have a limited effect. Four different forecasting methods are trained and tested on the same dataset with identical conditions. Performance analysis is made using well-known evaluation metrics. The results indicate that SARIMAX achieves the best forecasting performance for raw milk supply in Türkiye, followed by XGBoost.