Vector Autoregressive Model with Seasonal Indicator and Feed Forward Neural Network for Modeling Rainfall in Malang and Karangkates

Water is a basic need for the survival of every living thing. Rainfall is the main source of water availability. Lack of water supply can have a tremendous negative impact. On the other hand, heavy rainfall can cause flooding which has a bad impact. Accuracy in rainfall prediction is useful in crop planning strategies and flood and drought prevention. Research on rainfall used several models including Autoregressive (ARIMA), Seasonal ARIMA (SARIMA), Vector Autoregressive (VAR) and Feed Forward Neural – Network (FFNN). The purpose of this study is to establish a VAR with Seasonal Indicator and FFNN model for rainfall in Malang and Karangkates and compare the performance of the models. The novelty of this research is that we added the seasonal indicator variable to the VAR model as an exogenous variable and as an input to the feed forward neural network model. The best VAR model for rainfall in Malang and Karangkates is the first order VAR model (VAR(1)) with seasonal indicator variables. While the FFNN model for rainfall in Malang and Karangkates is the FFNN model with the tangent hyperbolic activation function and the number of units in the hidden layer is 15 and the inputs are seasonal indicator variables, rainfall in Malang the previous day and rainfall in Karangkates the previous day. The result of this study is VAR (1) with indicator variables model which is better than the VAR-NN with indicator variables model based on RMSE, especially on testing data.