Estimating sufficient sample sizes for approximate decision support queries

Abstract

Sampling schemes for approximate processing of highly selective decision support queries need to retrieve sufficient number of records that can provide reliable results within acceptable error limits. The k-MDI tree is an innovative index structure that supports drawing rich samples of relevant records for a given set of dimensional attribute ranges. This paper describes a method for estimating sufficient sample sizes for decision support queries based on inverse simple random sampling without replacement (SRSWOR). Combined with a k-MDI tree index, this method is shown to offer a reliable approach to approximate query processing for decision support.