A Novel Distance Measure for Data Vectors with Nominal Feature Values

摘要

The selection right distance measure is important for most machine learning algorithms. Euclidean distance is a commonly used distance measure in many methods due to simplicity of implementation. However the properties of problem domain are important thing, this selection must be done carefully. For example, most problems use data vectors with real-valued and nominal feature values. Euclidian distance produces reasonable results for real data, whereas it can not be said for nominal data. Hence in this study the new distance measure has been proposed for calculating distance between data vectors with nominal feature value. As the testing K-Means Clustering algorithm and the Mammographic Mass Data form UCI Repository have been used.