On self-organised regression curves
The practical effectiveness of the self-organising map (SOM) as an unsupervised clustering tool has been demonstrated in several socio-economic applications involving a large number of cases and/or variables. In a parametric SOM the family of operators to be mapped is continuous with respect to some u parameter vector. This paper discusses a specific form of PSOM which invokes a linear regression model; the input vector components are regarded as the p observations of a response variable. Three approaches to regression maps are discussed, the normal, posterior and prior methods. The prior method, in spite of slightly worse topology preservation achievement, performed reliably across the various experiments, neither requiring a twin learning session nor depending on the fitting power of the selected regression model.
Bibliographic Reference: Paper presented: ICANN '95 (International Conference on Artificial Neural Networks), Paris (FR), October 9-13, 1995
Availability: Available from (1) as Paper EN 39014 ORA
Record Number: 199510746 / Last updated on: 1995-07-07
Original language: en
Available languages: en