Data driven algorithm selection to predict agriculture commodities price

Abstract

Price prediction and forecasting are common in the agriculture sector. The previous research shows that the advancement in prediction and forecasting algorithms will help farmers to get a better return for their produce. The selection of the best fitting algorithm for the given data set and the commodity is crucial. The historical experimental results show that the performance of the algorithms varies with the input data. Our main objective was to develop a model in which the best-performing prediction algorithm gets selected for the given data set. For the experiment, we have used seasonal autoregressive integrated moving average (SARIMA) stack ensemble and gradient boosting algorithms for the commodities Tomato and Potato with monthly and weekly average prices. The experimental results show that no algorithm is consistent with the given commodities and price data. Using the proposed model for the monthly forecasting and Tomato, stack ensemble is a better choice for Karnataka and Madhya Pradesh states with 59% and 61% accuracy. For Potatoes with the monthly price for Karnataka and Maharashtra, the stack ensemble model gave 60% and 85% accuracy. For weekly prediction, the accuracy of gradient boosting is better compared to other models.