The PNC2 Cluster Algorithm is primarily intended for classification tasks. However, it is also possible to handle regression tasks. Therefore the continuous output is discretized using equidistant binning. The discretized output values can then be treated as different symbolic values and the algorithm can be used as ordinary. To compensate partly the loss of information, that results from the discretization, the output value of each cluster is re-calculated after the clustering is finished as the mean of the output values of all the data tuples merged within. More Info