THE STUDY OF OBJECTS CLUSTERING ALGORITHMS BASED ON SELF-ORGANIZING KOHONEN CARDS USING METHODS OF EXTRACTING FACTORS

We proposed algorithms for objects clustering based on self-organizing Kohonen maps using various methods of extracting factors (factor analysis: Principal Component Analysis, Maximum Likelihood Estimation, Principal Component Analysis based on Singular Value Decomposition). We performed experiments with different distance measures (Mahalanobis, Euclidean, squared Euclidean, Manhattan), various ways of neuron weights initialization (random, with choice of weight coefficients from a dataset). The computational experiments showed that the use methods of extracting factors in the SOM algorithm improves the accuracy of clustering in most cases. Moreover, clustering accuracy decreases with increasing number of homogeneous batches in a mixed lot.