Aldo Ramirez Arellano, J. Bory‐Reyes, L. Hernández-Simón
tlooto Summary
The results support the hypothesis that parametric algorithms are useful for datasets with numeric and nominal, but not for mixed, attributes; thus, four hybrid approaches are proposed.
Abstract
Themaingoalofthisarticleistopresentastatisticalstudyofdecisiontreelearningalgorithmsbased onthemeasuresofdifferentparametricentropies.Partialempiricalevidenceispresentedtosupport theconjecturethattheparameteradjustingofdifferententropymeasuresmightbiastheclassification. Here,thereceiveroperatingcharacteristic(ROC)curveanalysis,precisely,theareaundertheROC curve(AURC)givesthebestcriteriontoevaluatedecisiontreesbasedonparametricentropies.The authorsemphasizethattheimprovementoftheAURCreliesonofthetypeofeachdataset.Theresults supportthehypothesisthatparametricalgorithmsareusefulfordatasetswithnumericandnominal, butnotformixed,attributes;thus,fourhybridapproachesareproposed.Thehybridalgorithm,which isbasedonRenyientropy,issuitablefornominal,numeric,andmixeddatasets.Moreover,itrequires lesstimewhenthenumberofnodesisreduced,whentheAURCismaintainingorincreasing,thus itispreferableinlargedatasets. KEyWoRDS Classification, Data Mining, Decision Trees, Entropy Measures, Information Theory
Citation format
ARELLANO, Aldo Ramirez; BORY‐REYES, J.; HERNÁNDEZ-SIMÓN, L. Statistical entropy measures in c4.5 trees. INTERNATIONAL JOURNAL OF DATA WAREHOUSING AND MINING, 2018, 14: 1–14.