Comparing Biclustering Algorithms Using Data Envelopment Analysis to Choose the Best Parameters

Abstract

Biclustering method is one of the most important methods of the data mining techniques. Biclustering can be used to discover similar patterns in datasets, especially gene expression datasets or any datasets that can be presented as a matrix. Starting with Block Clustering algorithm in 1972 until now, a good number of the biclustering algorithms have been introduced. Each one of these algorithms was introduced to discover specific things in the data. In addition, each one of the introduced algorithms has some features differ from other algorithms. So far, we can say that there is no clear user manual to help in choosing the best algorithms. Another problem is to choose the parameters for each algorithm. Many works have been done aim to compare the biclustering algorithms according to some evaluating measures and no appropriate effort has been made to determine how to choose the best parameters under certain conditions. In this work, a two-stage comparison study is introduced. In the first stage, data envelopment analysis is used to choose the best parameters for each algorithm according to some measures. In the second stage, using the results of the first stage some of the introduced algorithms were compared according to the size and some different variance measures.