Predicting Potato Prices in Agra, UP, India: An H2O AutoML Approach

Abstract

The dynamics of the potato market in Agra, Uttar Pradesh, India, represent significant price volatility that affects stakeholders across the supply chain. This study addresses the critical need for accurate forecasting of potato price, which is utmost for optimising production, marketing strategies and inventory management. However, existing forecasting models often fail to provide the accuracy required for effective planning and resource allocation. This research aims to bridge this gap by investigating the potential of advanced predictive models to offer closer approximations of potato prices. Covering the period from January 1, 2006, to July 31, 2023, the methodology employed the H2O AutoML framework to identify and evaluate predictive models based on two distinct train-test split ratios, 80:20 and 70:30. The selection of the top 20 models for each configuration, assessed using the root mean square error, revealed the 70:30 split’s superior performance. Further analysis identified the top three models: stacked ensemble, gradient boosting machine and extreme gradient boosting, with the stacked ensemble model emerging as the optimal choice with forecasting errors ranging from 0.08 to 2.09% for daily prices of potato. This result illustrates the effectiveness of the stacked ensemble model in advancing strategic decision-making and resource distribution within the potato industry, with a notable improvement in the accuracy of price predictions contributing to more efficient and informed operational strategies.