ORANGE COMMODITY FORECASTING USING REGRESSION METHODS AND DATA SCIENCE

Abstract

The commodity market is crucial for the global economy, impacting sectors such as agriculture, energy, and metals. Its volatility stems from external factors like geopolitical instability, climate, and supply-demand variations, requiring constant risk assessment. This study evaluates two regression techniques — Ridge and LASSO — applied to orange price forecasting, considering polynomial expansion and the influence of training window size, with cross-validation (K-Fold) and metrics such as MSE and MAE. Results showed Ridge to be more robust against multicollinearity, while LASSO was more sensitive but benefited from calibration via validation. The appropriate choice of technique and preprocessing improves model generalization and supports safer decisions. The study highlights the importance of integrating statistical methods and emerging technologies in risk management within volatile markets.