A random forest-based machine learning algorithm for predicting onion whole sale price

The purpose of this study is to estimate and predict onion wholesale price volatility using statistical and machine learning algorithms. Traditional models like ARIMA and GARCH were compared against advanced machine learning techniques like SVM, LSTM, and CNN, as well as hybrid approaches like ACF-LSTM, RF-LSTM, ACF-CNN, and RF-CNN. The performance of these models was assessed using RMSE, MAPE, and MAE. The results show that hybrid models, specifically RF-LSTM and RF-CNN, outperformed classic statistical models. Specifically, RF-LSTM and RF-CNN outperformed the statistical models by 62% in RMSE, 76% in MAPE, and 71 % in MSE for the training set, and 76% in RMSE, 78% in MAPE, and 74% in MSE for the testing set. These results highlight the superiority of hybrid models in describing the intricate non-linear patterns in the volatility of onion prices, providing agricultural commodities markets with more reliable and precise forecasting tools. The findings are expected to inform better decision-making processes for stakeholders across the onion supply chain, from farmers to policymakers.