Study on forecasting of Indian onion export price using machine learning techniques-based hybrid models

Abstract

The onion market is profoundly shaped by global trade dynamics, rapid technological advancement, and shifting consumer demand. Within this global context, India stands out as the world's largest producer and second-largest exporter of onions. Despite this prominent position, the nation's economy and export potential are threatened by significant price volatility. Consequently, accurate price predictions are crucial for farmers, policymakers, and the government to make informed decisions. Unfortunately, agricultural datasets often exhibit nonlinearity and non-stationarity, making predictions challenging. To address this, hybrid models are proposed, which combine multiple models instead of relying on individual ones. Specifically, this article compares the performance of individual models (ARIMA, TDNN, SVR) against a suite of hybrid models (EMD-ARIMA, EMD-TDNN, EMD-SVR, EEMD-ARIMA, EEMD-TDNN, EEMD-SVR). Monthly onion price data from 2008 to 2021 is used, and the dataset is decomposed into six independent intrinsic modes and one residue, revealing price volatility patterns. The decomposed components are forecasted using conventional and machine learning methods, and the forecasts are aggregated to produce a final prediction. Empirical evaluation demonstrates that the EEMD-SVR model achieves superior predictive accuracy, with the lowest RMSE (405.09) and MAPE (14.93%). This highlights its effectiveness in modeling the complex, nonlinear, and non-stationary dynamics of agricultural prices.