This research is concerned with handling randomness in time series data, especially elimination of random movement to make the data predictable using forecasting models. The noise data extracted two sets of time series data for block travel time showed higher random property, and that the de-noised travel time data showed much higher regularity than the original data, based on Hurst exponent and Kaplan's statistics. In the state-space model (AR-Type) based forecasting, very accurate results were obtained for the two sets of times series data. As a result, it is shown that accuracy in forecasting and deterministic regularity can be enhanced in the de-noising process, even with highly irregular time series data. For the reason, de-noising process can be usefully employed as a pre-processing step in developing a time series forecasting. (A*)
Abstract