A New Cluster Validity Index for Stock Clustering Based on Efficient Frontier

Clustering is an unsupervised learning method to discover meaningful information by grouping similar objects together. It is a great challenge to valuate the results of stock clustering. In this paper, we propose a specific index IBEF(Index Based on Efficient Frontier) to evaluate the results of stock clustering based on the concept of efficient frontier. IBEF is defined by the difference between two efficient frontier curves. One curve is built by all stocks and the other curve is built by center stock of each cluster. If the clustering result is good, the two curves are close to each other and IBEF value will be small. Our experiments on different correlation coefficients and clustering methods show that IBEF is a proper validity index comparing with other indexes.