Mi SciELO
Servicios Personalizados
Revista
Articulo
Indicadores
- Citado por SciELO
Links relacionados
- Similares en SciELO
Compartir
Revista Cubana de Ciencias Informáticas
versión On-line ISSN 2227-1899
Resumen
GONZALEZ, Héctor; SANTOS, Gabriela; CAMPOS, Frank y MORELL PEREZ, Carlos. Evaluation of KNN-SP algorithm for multi-target prediction problems. Rev cuba cienc informat [online]. 2016, vol.10, n.3, pp.119-129. ISSN 2227-1899.
ABSTRACT This paper aims to develop a statistical evaluation of KNN algorithm for multi-target prediction problems. The cross-validation procedure is used to set the parameter and inverse distance as weigth to generate 4 variants of KNN for muti-target prediction. The study of statistical methods for comparison of multiple classifiers identified the non-parametric Friedman test as one of the most used in the testing of machine learning algorithms. In the experimental results, it employ 12 standards dataset and the metric aRRMSE. With application of the Nemenyi post-hoc and Friedman tests showed that cross-validation procedure applied in IBKMTRCV and IBKMTRCVW is significantly better than the variants that do not use this procedure. The values of the average ranking of the Friedman test, show that IBKMTRCVW algorithm returns the best results.
Palabras clave : machine learning; comparisons of multiple classifiers; KNN; Multi-target prediction.