An algorithm for designing a pattern classifier by using MDL criterion

The algorithm for designing a pattern classifier, which uses MDL criterion and a binary data structure, is proposed. The algorithm gives a partitioning of the space of the K-dimensional attribute and gives an estimated probability model for this partitioning. The volume of bins in this partitioning is asymptotically upper bounded by /spl Oscr/((log N/N)/sup K/(K+2/)/sup )/ for large N in probability, where N is the length of training sequence. The redundancy of the code length and the divergence of the estimated model are asymptotically upper bounded by /spl Oscr/(K(log N/N)/sup 2/(K+2/)/sup )/. The classification error is asymptotically upper bounded by /spl Oscr/(K/sup 1/2/(log N/N)/sup 1/(K+2/)/sup )/.