A modified condensed nearest neighbour rule using the symbolic approach

The Condensed Nearest Neighbour (CNN) method chooses samples randomly. This results in: retention of unnecessary samples; occasional retention of internal rather than boundary samples. A modification of the CNN method is presented which will overcome these disadvantages by considering only points close to the decision boundary. This is achieved by making use of the symbolic approach. In conventional data analysis, the samples are described by feature vectors of numeric type. However, more generalised description of samples contain interval type of features and qualitative features. Such a description of samples constitute symbolic data. A modified new similarity measure which takes into consideration the position, span and content of symbolic objects is proposed. The similarity measure used is of new type. The modified similarity measure is used for obtaining the samples of the condensed set. The performance of the proposed modified CNN rule is illustrated by an example. The results obtained by the proposed method are an improvement over the other existing methods.< >