Free k-means clustering software

It includes k-Means and Hierarchical Clustering. PermutMatrix provides data visualizations with clustering and seriation analysis. It supports hierarchical clustering. It is built on python and specifically NumPy, SciPy and matplotlib, and supports many clustering methods including k-Means, affinity propagation, spectral clustering, Ward hierarchical clustering, agglomerative clustering hierarchical , Gaussian mixtures and Birch clustering.

KNIME is a general purpose data mining platform with over different operators. Orange is a relatively easy to use data mining platform with support for hundreds of operators. NeuroXL Clusterizer , a fast, powerful and easy-to-use neural network software tool for cluster analysis in Microsoft Excel. Neusciences aXi.

Download available. Also for OEM. Viscovery explorative data mining modules , with visual cluster analysis, segmentation, and assignment of operational measures to defined segments. This method minimizes an objective function by swapping objects from one cluster to another. Beginning at a random starting configuration, the algorithm proceeds to a local minimum by intelligently moving objects from one cluster to another.

When no object moving would result in a reduction of the objective function, the procedure terminates. Unfortunately, this local minimum is not necessarily the global minimum. To overcome this limitation, the program lets you rerun the algorithm using several random starting configurations and the best solution is kept. This algorithm also attempts to minimize the total distance D formula given above between objects within each cluster. The algorithm proceeds through two phases.

In the first phase, a representative set of k objects is found. The first object selected has the shortest distance to all other objects. That is, it is in the center.

An addition k-1 objects are selected one at a time in such a manner that at each step, they decrease D as much as possible. In the second phase, possible alternatives to the k objects selected in phase one are considered in an iterative manner.

At each step, the algorithm searches the unselected objects for the one that if exchanged with one of the k selected objects will lower the objective function the most. The exchange is made and the step is repeated. These iterations continue until no exchanges can be found that will lower the objective function. Note that all potential swaps are considered and that the algorithm does not depend on the order of the objects on the database.

Fuzzy clustering generalizes partition clustering methods such as k-means and medoid by allowing an individual to be partially classified into more than one cluster. In regular clustering, each individual is a member of only one cluster.

Suppose we have K clusters and we define a set of variables that represent the probability that object i is classified into cluster k. In partition clustering algorithms, one of these values will be one and the rest will be zero. This represents the fact that these algorithms classify an individual into one and only one cluster. In fuzzy clustering, the membership is spread among all clusters. The probability of each object to be in each cluster can now be between zero and one, with the stipulation that the sum of their values is one.

We call this a fuzzification of the cluster configuration. It has the advantage that it does not force every object into a specific cluster. It has the disadvantage that there is much more information to be interpreted. The algorithm partitions the data into two or more clusters and performs an individual multiple regression on the data within each cluster.

It is based on an exchange algorithm described in Spath. This algorithm is fairly simple to describe. The number of clusters, K, for a given run is fixed.