Efficient quantisation of continuous valued data for machine learning

The paper presents a new algorithm for quantisation of continuous-valued attributes for machine learning applications. The algorithm carries out discretisation in such a way that the quantisation levels obtained embed the local and the global information contained in the data set. The paper defines some terms that give a measure of the local and global information present in the data set. The quantisation procedure is carried out by minimising the disorder measure defined in the paper A low disorder measure ensures the embedding of local and global information in the quantisation levels. The algorithm is used as a front end for a number of inductive learning algorithms and tested on a bench marking data set. The paper compares the results of the suggested algorithm with that of RULES2 and RULES3 algorithms.