A computational method for clustering evaluation in toxic activity study

A validation procedure to evaluate the optimal number of clusters in a toxic activity data set using automatic classification by crisp partitioning clustering is here presented. The clustering procedures used are based on either the Kohonen, SOM, and kmeans joined algorithms. The clustering validation method consists in the use of an index whose task is to choose the clusters optimal number. The here used index is the Mutual Information. The index providing the indication regarding the optimal number of clusters allows to reach more reliable clustering results. This procedure has been applied to the study of the phytotoxic activity of some species of Salvia genus plants. To reduce experimentation costs, only the most significant species has to be involved in trials; hence the importance to recognize the different levels of Salvia species phytotoxicity, and to provide a suitable classification, performed by ranking methods.