Pemodelan prediksi total hujan pada musim hujan menggunakan jaringan saraf tiruan dan support vector regression
Artificial neural network and support vector regression to predict total rainfall in wet season
Abstract
The Artificial Neural Network (ANN) and Support Vector Regression (SVR) models were developed to predict the total of rainfall in wet season in Indramayu. Onset Data, length of rainy season and monthly Southern Oscillation Index value on August, October and February are used as the input of the models. ANN employed in this study was multilayer perceptron with neuron hidden layer as many as 10, 30 and 50, and trained with gradientdescent backpropagation algorithm. SVR employed with three kinds of kernel functions, linear, polynomial and radial basis function. Models trained with three scenarios length of training data i.e. 15, 20 and 25 periode/ years. This research compared perfomance of two models by roort of mean squared error (RMSE), mean percent of error (MAPE) and correlation coefficient (R) values. The optimal performance of ANN model is mean of percent error 22,70%, root of mean squared error 231,32, and correlation 0,46 , resulted from model with 30 units hidden neuron and trained by 20 training data. SVR models showed better performance then ANN models. SVR model with linear kernel trained by 20 training data show the best performace with root of mean squared error 120,60, correlation 0,86 and mean of percent error 9,09%. Fenomena iklim merupakan salah satu faktor yang sulit dikendalikan dan berpengaruh pada produktifitas tanaman pangan. Salah satu upaya untuk mengurangi resiko kerugian ekonomi dari hilangnya investasi pada proses penanaman maupun gagal panen adalah dengan cara mendeteksi dini fenomena iklim di masa mendatang. Kondisi ini mendorong dilakukannya prediksi iklim, khususnya prediksi curah hujan. Total hujan pada musim hujan memberikan gambaran kondisi ketersediaan air pada musim hujan dan musim berikutn