A Hybrid Model for Monthly Time Series Forecasting

This study aims to propose a hydrological model for estimating the future value for monthly river flow. The proposed model was constructed by combining three components: i.e. Discrete Wavelet Transform (DWT), Principal Component Analysis (PCA) and Least Square Support Vector Machine (LSSVM). The first two components, i.e. the wavelets and the PCA, are meant for preparing input data. Wavelets were employed to obtain a certain level of data decomposition, and in this case, a three level decomposition was employed. The output from the wavelets was given to PCA. This component simply picks up the important components from the given data, i.e. it addresses the issues relating to the dimensionality of the data. For approximating the desired value, LSSVM was employed for training, using the data derived from Wavelets and PCA models. For testing stability and reliability of the proposed model monthly data from two Pakistani rivers was collected. The reliability was measured by employing well known reliability measuring methods, i.e. Root Mean Square Error (RMSE), Mean Absolute Error (MAE) and Coefficient of Correlation (R). All performance measuring methods concluded that the proposed model is stable, reliable and produced an appreciative level of accuracy