Pemodelan Jaringan Syaraf Tiruan untuk Memprediksi Awal Musim Hujan Berdasarkan Suhu Muka Laut

Abstract

Anjatan-Indramayu is the one of agriculture regions in Indonesia. The success or failure of the harvest each year depends on water availability in the region. Therefore we need an accurate method to predict the beginning of rainy season. The method used to predict in this research is artificial neural network (ANN) backpropagation. Result of ANN prediction accuracy is measured by R2 and RMSE. This research uses ECHAM4p5_CA sea surface temperature (SST) is one of the sea surface temperature models with the months of June, July and August. The domain of SST is selected based on the correlation of 5% and 10% for each of the month June, July, and August. This research uses ANN architecture with two parameters: neuron hidden (NH) and learning rate (LR). The number of hidden neurons used in this research are 5, 10, 20, and 40, and the learning rates are 0.3, 0.1, and 0.01. The best prediction result correlation of 5% using the ANN is for June with R2 is 51% and RMSE 3.03 at NH 10 and LR 0.01, July with R2 is 48% and RMSE 3.39 at NH 20 and LR 0.1, and August with R2 is 75% and RMSE 2.51 at NH 40 and LR 0.01. The best prediction result correlation of 10% using the ANN is for June with R2 is 44% and RMSE 3.32 at NH 5 and LR 0.3, July with R2 is 42% and RMSE 3.42 at NH 10 and LR 0.1, and August with R2 is 71% and RMSE 3.37 at NH 20 and LR 0.01. The conclusion from this research is the neuron hidden and learning rate with different values affect R2 and RMSE.