Forecasting Banana Harvest Yields using Deep Learning

Abstract

In agriculture sector, forecasting yields such as in banana harvest is essential for extensive planning for a sustainable production. However, studies have shown that agricultural problems like forecasting yields remain difficult due to the lack of the necessary and optimal infrastructures. Recently, deep-learning (DL) approach like the recurrent neural network- long short term memory (RNN-LSTM) has been used as a new alternative model in time series (TS) forecasting. In this paper, using the banana harvest data from agrarian reform beneficiary (ARB) cooperative in Davao del Norte, Philippines, proved that RNN-LSTM has the capability in forecasting harvest yields over conventional-based model like the famous autoregressive integrated moving average (ARIMA). Using the same set of training and testing data, experiment exhibits that RNN-LSTM obtains 43.69 in terms of root-mean-squared-error (RMSE) and ARIMA obtains 64.11respectively. This means that RNN-LSTM outperforms the ARIMA model with 32.31 percent reduction in error rates. Since there is no proof that RNN-LSTM has been used with the same agricultural problem domain, therefore, there is no standard available with regards to the level of error reduction in the forecast. Moreover, as enhancement to the performance of the model, the use of multiple LSTM layers joined together and the use of regularization technique like dropout are suggested for exploration. Furthermore, extending and embedding this approach to a web-based application is the future plan for the benefit of the medium scale banana growers of the region for efficient and effective decision making and advance planning.