A Hierarchical RF-XGBoost Model for Short-Cycle Agricultural Product Sales Forecasting.

Area of Science:

• Agricultural Economics
• Data Science
• Supply Chain Management

Background:

• Short-cycle agricultural product sales forecasting is crucial for minimizing food waste by aligning supply with demand.
• Forecasting accuracy is challenged by volatile and discontinuous sales data due to uncertain factors.
• Existing models often struggle with the inherent complexities of agricultural market dynamics.

Purpose of the Study:

• To develop and evaluate a novel hierarchical prediction model for enhanced short-term agricultural product sales forecasting.
• To improve demand prediction accuracy, thereby reducing food waste and optimizing the agricultural supply chain.
• To address the volatility and discontinuity in sales data through an advanced modeling approach.

Main Methods:

• A hierarchical model combining Random Forest (RF) and Extreme Gradient Boosting (XGBoost) was developed.
• The first layer uses RF with Grey Relation Analysis (GRA) for initial predictions and residual extraction.
• The second layer employs XGBoost on residual clustering features for refined forecasting.

Main Results:

• The proposed RF-XGBoost model demonstrated superior performance compared to standalone RF and XGBoost.
• Achieved a 10% and 12% reduction in Mean Absolute Percentage Error (MAPE) over RF and XGBoost, respectively.
• Showcased a 22% and 24% increase in the coefficient of determination (R²), indicating higher prediction accuracy.

Conclusions:

• The hierarchical RF-XGBoost model significantly enhances the precision of short-term agricultural sales forecasting.
• The model's effectiveness was validated across diverse agricultural products, demonstrating broad applicability.
• Implementation offers substantial benefits for supply chain optimization and food waste reduction.