Forecasting Trend-Seasonal Data Using Nonparametric Regression with Kernel and Fourier Series Approach

Abstract

Recently, forecasting time series data with trend and seasonal or trend-seasonal combinations with time series forecasting methods that are often used, are bound by assumptions that must be reached. If it does not reach the assumptions that exist, the forecasting process becomes longer. This study provides an alternative approach used for time series data forecasting that has trend-seasonal combination pattern using nonparametric regression. Some nonparametric regression approaches such as the kernel and the Fourier series can be done by considering the predictor as the time scale for a regular period. For the same data, using Nadaraya–Watson kernel approach and Fourier series in nonparametric regression gives different results. The result of prediction using nonparametric regression with the Fourier series approach is closer to the original data, when compared to the kernel approach. For each oscillation parameters inputted, nonparametric regression with the Fourier series approach always provides smaller MSE results than MSE in every bandwidth for Nadaraya–Watson kernel approach.