Métodos de clasificación en minería de datos meteorológicos.

Abstract

One of the colectivamente of data mining is classification, which aims to classify a variable within one of the categories of a class. In this work, meteorological variables of the Cumandá station were considered. In order to determine the appropriate model for the data set, the Naive Bayes, CN2 Rule Induction, K-NN, Tree and Random Forest classification models, as well as Cross validation, Random sampling, Leave one out and Test on train data, that modify the parameters associated with the classifier, were applied. Orange Performance software was used to calculate performance measures, Classification Accuary, Global Accuracy and Sensitivity . It was concluded that the classifiers Naive Bayes , CN2 Rule Induction and K-NN presented values higher than 75% of correctly classified instances. The decision tree and the Random Forest exceeded 80%. Regarding the methods that allow to modify the parameters associated to the classifier, it was determined that Cross-validation presented better results in all the applications. The highest precision is reached in the classifier random forest with 83.9% applying cross-validation, followed by simple random sampling with 83.1% of true positives among the cases classified as positive.