Data Ranking Based on Spatial Partitioning

Abstract

In applications where preferences are sought it is desirable to order instances of important phenomenon rather than classify them. Here we consider the problem of learning how to order instances based on spatial partitioning. We seek statements to the effect that one instance should be ranked ahead of another.

A two-stage approach to data ranking is proposed in this paper. The first learns to partition the spatial areas using the largest irregular area to represent the same rank data recursively, and gets a series of spatial areas (rules). The second stage learns a binary preference function indicating whether it is advisable to rank one instance before another according to the rules obtained from the first stage. The proposed method is evaluated using real world stock market data set. The results from initial experiments are quite remarkable and the testing accuracy is up to 71.15%.