New Assessment Criteria for Clustering Algorithms

Clustering algorithms offer several advantages over a manual grouping processes. Yet, the initial and random guess of most clustering algorithms, along with noise and outliers, affect the reliability of their results. In this paper, new clustering performance measures (CPM) for assessing the reliability of a clustering algorithm are proposed. In this paper, two parameters are used to define clustering performance measures - the first is the validation measure, which is used for determining how well the algorithm works at a given set of parameter values, and the second is a repeatability measure, which is used for studying the effect of initial conditions on the clusters membership. Furthermore, these CPMs can be used to evaluate clustering algorithms. Two different types of real-world data are used for such an evaluation procedure. The first is a communications signal data set representing one modulation scheme under noise condition, and the second is a breast cancer data set