Heterogeneous distance functions for prototype rules : influence of parameters on probability estimation

摘要: An interesting and little explored way to understand data is based on prototype rules (P-rules). The goal of this approach find optimal similarity (or distance) functions position prototypes which unknown vectors are compared. In real applications frequently involve different types attributes, such as continuous, discrete, binary or nominal. Heterogeneous distance that may handle diverse information usually probability measure, the Value Difference Metrics (VDM). For continuous attributes calculation probabilities requires estimations density functions. This process careful selection several parameters have important impact overall classification accuracy. paper various heterogeneous function VDM measure presented, among them some new estimation. Results many numerical experiments with presented artificial datasets, quite simple P-rules for databases extracted.