Full Program »
A Comparative Study of Classifier Algorithms For Recommendation of Banking Products
Currently, the processing and storage capacities of computers have facilitated the handling of large volumes of data. Robust classification models are used by organizations in various sectors, but the algorithms are not efficient with all data sets; in fact, high-volume data can significantly change the most promising algorithms. In this work, a comparative study of classification algorithms was developed; To do this, we use a high-volume data set related to banking products purchases. We selected eight important algorithms from the literature in the first stage, but only five reached a second stage after the tree-based algorithms presented memory overflow problems. The criteria for quality evaluation were based on AUC(Area under the curve) measurements and training and testing time; in this investigation, the MARS algorithm was shown to be slightly superior to SVM in classification quality; while in execution times, both for training and for testing, MARS reached lower times. Both the MARS and SVM algorithms could be used as classifiers in this data set; however, the model construction costs should be considered when fractional construction methods are not used.