Reducing the Computational Cost of Computing Approximated Median Strings

The k-Nearest Neighbour (k-NN) rule is one of the most popular techniques in Pattern Recognition. This technique requires good prototypes in order to achieve good results with a reasonable computational cost. When objects are represented by strings, the Median String of a set of strings could be the best prototype for representing the whole set (i.e., the class of the objects). However, obtaining the Median String is an NP-Hard problem, and only approximations to the median string can be computed with a reasonable computational cost. Although proposed algorithms to obtain approximations to Median String are polynomial, their computational cost is quite high (cubic order), and obtaining the prototypes is very costly. In this work, we propose several techniques in order to reduce this computational cost without degrading the classification performance by the Nearest Neighbour rule.