USING DATA MINING TO IDENTIFY A CONSISTENT SET OF EXPERTS FOR TIMES SERIES SALES FORECASTING

摘要

Combining forecasts from different models has shown to perform better than single forecasts in most times series. Recent research in Rank based combining methods have yielded interesting results. The number of methods used by these rank based techniques number almost a million. In the final analysis only few 'K' methods perform well for a given series. This paper presents a novel procedure based on the classical association rule mining algorithm to find the best 'K' methods for a given time series.