Background Accurate forecasting of agricultural commodity prices is crucial for maintaining economic stability and informing data-driven policy interventions in emerging economies, such as India. Price volatility in key spices, such as pepper and turmeric, can significantly impact producers, consumers, and market stakeholders, underscoring the need for robust predictive models that capture both short-term fluctuations and long-term dependencies. Methods This study develops an integrated forecasting framework that combines Machine Learning (Random Forest), Deep Learning (LSTM), and Econometric (VECM) approaches to analyse the dynamic behaviour of Indian pepper and turmeric prices. The models incorporate major macroeconomic determinants, including GDP, Consumer Price Index (CPI), exchange rate, gold price, interest rate, trade volume, and foreign institutional investments (FII), to capture both non-linear and long-term relationships. Model performance was evaluated using RMSE, MAE, and symmetric MAPE (sMAPE) metrics, alongside SHAP-based feature explainability analysis. Results The findings reveal that Random Forest delivers the most robust predictive accuracy overall, especially for Turmeric, while LSTM achieves slightly lower forecast errors for Pepper. Both machine-learning models substantially outperform the VECM in short-term price forecasts. Feature importance and SHAP analyses identified NIFTY50, GDP, CPI, exchange rate, interest rate and gold prices as key drivers of spice price movements. Conclusions Integrating machine learning, deep learning, and econometric models enhances the robustness and interpretability of commodity price forecasting. The study provides empirical evidence that macroeconomic variables significantly influence spice price dynamics, offering a hybrid framework that can support policymakers, traders, and researchers in mitigating market risks and designing more effective agricultural price stabilisation strategies.
Vaishnavi et al. (Tue,) studied this question.