Abstract: For the sensitivity of noise and outliers data in the typical partitioning clustering algorithm, a clustering algorithm based on data dispersion was proposed. The data dispersion was defined and introduced to a non-Euclidean distance. The similarity metric was established, and the data clustering was realized. The optimal clustering number was obtained by the validity function based on improved partition coefficient. Then the proposed clustering algorithm was applied to quality index model in slashing process. A size add-on quality index model was built by radial basis function neural networks. The node number of hidden layer was determined and the center of radial basis function was obtained by the proposed clustering algorithm. The empirical result shows that the clustering result is insensitive to noise and outliers data, and the accuracy of size add-on quality index model is higher.

Key words: quality index model, clustering, data dispersion, non-Euclidean distance, textile slashing process