There are applications that require ordered instances modeled by high dimensional vectors. Despite the reasonable quantity of papers on the areas of classification and clustering and its crescent importance, papers on ranking are rare. Usual solutions are not generic and demand expert knowledge on the specification of the weight of each component and, therefore, the definition of a ranking function. This paper proposes a generic procedure for ranking, based on 1D self-organizing maps (SOMs). Additionally, the similarity metric used by SOM is modified and automatically adjusted to the context by a genetic search. This process seeks for the best ranking that marches the desired probability distribution provided by the specialist expectation. Promising results were achieved on the ranking of data from blood banks inspections.
[1]
Teuvo Kohonen,et al.
The self-organizing map
,
1990
.
[2]
Yoram Singer,et al.
Learning to Order Things
,
1997,
NIPS.
[3]
Laurene V. Fausett,et al.
Fundamentals Of Neural Networks
,
1993
.
[4]
Teuvo Kohonen,et al.
Self-Organizing Maps
,
2010
.
[5]
John G. Taylor,et al.
On the Ordering Conditions for Self-Organizing Maps
,
1995,
Neural Computation.
[6]
Miguel Á. Carreira-Perpiñán.
One-to-many mappings, continuity constraints and latent variable models
,
1999
.