Abstract
This research uses four classification algorithms in standard and boosted forms to predict members of a class for an online community. We compare two performance measures, area under the curve (AUC) and accuracy in the standard and boosted forms. The research compares four popular algorithms Bayes, logistic regression, J48 and Nearest Neighbor (NN). The analysis shows that there are significant differences among the base classification algorithms—J48 had the best accuracy. Additionally, the results show that boosted methods improved the accuracy of logistic regression. ANOVA was used to detect the differences between the algorithms; post hoc analysis shows the differences between specific algorithms.
Recommended Citation
Sibona, Christopher and Brickey, Jon, "A Statistical Comparison of Classification Algorithms on a Single Data Set" (2012). AMCIS 2012 Proceedings. 2.
https://aisel.aisnet.org/amcis2012/proceedings/ResearchMethods/2
A Statistical Comparison of Classification Algorithms on a Single Data Set
This research uses four classification algorithms in standard and boosted forms to predict members of a class for an online community. We compare two performance measures, area under the curve (AUC) and accuracy in the standard and boosted forms. The research compares four popular algorithms Bayes, logistic regression, J48 and Nearest Neighbor (NN). The analysis shows that there are significant differences among the base classification algorithms—J48 had the best accuracy. Additionally, the results show that boosted methods improved the accuracy of logistic regression. ANOVA was used to detect the differences between the algorithms; post hoc analysis shows the differences between specific algorithms.