Forecasting cashew production in India using a hybrid machine learning framework with STL decomposition, ensemble...

Area of Science:

• Agricultural Economics
• Data Science
• International Trade

Background:

• India's cashew production and Cashew Nut Shell Liquid (CNSL) exports are critical to the global market.
• Accurate forecasting and trade network analysis are essential for strategic decision-making in the agri-business sector.

Purpose of the Study:

• To develop an integrated analytical framework for forecasting India's cashew production and analyzing CNSL export trade dynamics.
• To identify key trading nodes and assess trade concentration within India's global CNSL network.
• To provide a decision-support system for policymakers and exporters.

Main Methods:

• A hybrid machine learning model combining Seasonal-Trend Decomposition using Loess (STL) with Random Forest and Gradient Boosting Machines for production forecasting.
• Network topology analysis using centrality measures (Degree, Closeness, Betweenness, Eigenvector, PageRank) and a novel Source-Importer Ratio for trade network characterization.
• Implementation of a novel rolling STL decomposition and an expanding window cross-validation for robust time series analysis.

Main Results:

• Gradient Boosting models achieved superior production forecasting accuracy (R²=0.988, MAPE=3.6%, RMSE=45.8 MT).
• India was identified as the dominant hub in the CNSL trade network, exhibiting maximal degree centrality and PageRank.
• Network analysis revealed a star-like topology with significant disparities in trade influence among trading partners.

Conclusions:

• The integrated framework provides high-accuracy forecasts and network diagnostics for India's cashew and CNSL sectors.
• Policy recommendations focus on enhancing supply chain resilience, mitigating trade risks, and promoting export diversification.
• The study emphasizes the importance of data-driven insights for strategic planning in the agricultural export sector.