A Long Short Term Memory Implemented for Rainfall Forecasting

Abstract

The prediction and its accuracy of the rainfall is needed due to it would be affected to the various areas of life, such as feasibility aircraft departures and, in general issue, is climate change. This paper aimed to apply a Long Short Term Memory (LSTM) approach to get accurate rainfall forecasting. Also, the LSTM accuracy would be compared to BPNN (Backpropagation Neural Network) algorithm. In this research, LSTM architecture used a hidden layer of 200, a maximum epoch of 250, 1 gradient threshold, and learning rates of 0.005, 0.007, and 0.009. Then, standardize data was used gamma γ of 1.05. Then, the BPNN architectures of [2-50-10-1, epoch 250] have been explored. The accuracy performance is measured by the root means square error (RMSE). The experimental results showed that the LSTM had produced a good accuracy than BPNN, with the value of RMSE was 0.2367 and 0.1938. It means that the forecast accuracy of the LSTM approach outperformed the BPNN to predict the rainfall. This finding would be useful for the climatology station to develop a forecsat rainfall application-based artificial intelligence.