Nearest neighbor method in time series forecasting

Abstract

The objective of knowledge discovery and data mining is to support decision-making through the effective use of information. To an increasing extent over the past decade, software learning methods including neural networks and case based reasoning(CBR) have been used for prediction in financial markets and other areas. CBR has been applied to many tasks, including the prediction. By extending the notion of an elementary case and using multiple neighbors, case reasoning can at times outperform neural networks, which perhaps represents the most widely used learning technique in practice. This thesis shows that the nearest neighbor method has a limitation on applying to nonstationary time series forecasting and suggests an alternative nearest neighbor method to predict the nonstationary time series by adopting the process of model identification in ARIMA, and illustrates that this method can be used effectively in forecasting the sales data which is nonstationary time series.