Classification Model Supervised Using the Heaviside Function

Andrés García Floriano


In this paper, the theoretical foundations of a new classification model which is based on the associative approach of Pattern Recognition: Heaviside’s Classifier. As its name suggests its both phases, learning and classification, are based on the Heaviside’s function. The effectivity of the proposed model can be verified by the results of a comparative study where the classifier was tested against other seven pattern recognition models on 20 datasets. Experimental results indicate that the model is competitive in the state of the art. It is noteworthy that with one dataset, our classifier achieved the 100% of performance, validated with 10-fold cross-validation, while in its worst performance it achieved a little above of 50%. The obtained results were validated by the Wilcoxon non parametric test, which provides statistical certainty to the results of the performance comparison between models.


Pattern classification, supervised learning, heaviside function, non parametric statistical test

Full Text: PDF (Spanish)