Ecuador Agricultural Product Price Forecast: A Comparative Study of Deep Learning Models

The application of forecasting techniques in the agriculture industry started with a commodity prediction almost a century ago. However, currently, the same application is not explored in the same field. For instance, in Ecuador, farmers have to suffer the volatility of prices of agriculture products during all the growing stages since they do not count on any forecasting method for preventing future events. Therefore, this work aims to reduce the gap of knowledge by presenting the implementation of five deep learning algorithms which forecast weekly and monthly prices of avocado, red onion, and cucumber from the wholesale market of Ibarra city in Ecuador. Results have shown that single models are still suitable for forecasting, although, the best performance comes from compound models such as Conv-LSTM-MLPs. Likewise, with proper hyperparameter tuning, the last model showed an error reduction (MAE) of 23% for weekly avocado prices.