STUDY OF MULTIPLE SEASONAL AUTOREGRESSIVE INTEGRATED MOVING AVERAGE SUBSEQUENCES AGGREGATE LONG-TERM TIME SERIES MODEL FOR FLOOD PREDICTION BASED ON THE SEASONAL RAINFALL DATA

Author

Sri Supatmi, S.Kom.,M.T., D.Sc Irfan Dwiguna Sumitra, S.Kom., M.Kom. Ph.D

Abstrak

This paper proposed a new prediction method for flood vulnerability during the wet season in Indonesia. This method called Multiple Autoregressive Integrated Moving Average Subsequences Aggregate Long-Term Time Series model or MSARIMASA model. The long-term time series data and divided into data actual-sampling for experiment and data fits-sampling for evaluating. MSARIMASA model built from the aggregate data actual-sampling and subdivided into training long-term time-series data as well as authorizing longterm time-series data to reduce the effect in the rainfall prediction based on time series employing the aggregate method by its period. A fixed number of subsequence patterns generated and fitting well using SARIMA models. It is also verified by the mean absolute percentage error (MAPE) and mean forecast error (MFE) to identify the best-fitted model of MSARIMASA. The step of predicting future rainfall in the aggregate grouped employing the MSARIMASA models needs to determine the subsequence with the best fitted SARIMA model. The disaggregates process to spread this group value to the future rainfall using a ratio from the previous rainfall data. The MSARIMASA model was compared using MAPE and MFE with the SARIMA model and the ARIMA model. The results for the proposed model indicated better MAPE and MFE than the SARIMA and ARIMA models