Forecasting Rainfall in Surabaya Using the Singular Spectrum Analysis Method Soehardjoepri, Ulil Azmi, Ika Safitri, Ivan
Departement of Actuarial Science, Institut Teknologi Sepuluh Nopember, Surabaya, Indonesia
Abstract
Singular Spectrum Analysis (SSA) is a time series analysis method that uses a non-parametric approach. This study aims to determine the model and predict rainfall in the city of Surabaya. The SSA process begins with decomposition consisting of embedding and singular value decomposition. Then a reconstruction process is carried out, which consists of eigentriple grouping and diagonal averaging. The determination of eigentriple grouping is based on changes in the value of the eigenvectors, which do not start much different between each eigenvector. This study will compare various window length (L) values. The best model obtained is with L = 72, MAD value is 151.9903, MSE is 32340.72, and sMAPE is 0.2679602. These results indicate that the best model will be obtained when using a maximum L, which is half of the total data or as much as 72. The results of rainfall forecasting using SSA cannot be used in the long term.
