Intermittent Data Forecasting using Kernel Support Vector Regression

Abstract

Forecasting involves making future estimates. Forecasting methods are commonly employed to predict stock prices, monetary distribution, and weather conditions. To generate accurate forecasts, it is crucial that the data used is consistent, comprehensive, and unchanging. Some data can be readily predicted, while some poses a considerable challenge. An illustration of this is found in discontinuous data, which is notably hard to forecast. Discontinuous data is marked by frequent instances of zero values due to sporadic events. For instance, when tracking the sales of aircraft or other products, sales do not transpire daily, causing recorded data to often register as zero. Various techniques have been explored to handle this kind of data. In this particular study, the chosen method is support vector regression. This method is capable of predicting discontinuous data with a quality level of 1.004, which is lower than traditional approaches like exponential smoothing.