Construction features of scoring models based on DEDUCTOR analytical platform

• Main Authors: Lagerev D.G., Bondareva I.V.
• Format: Article
• Language: deu
• Published: Academician I.G. Petrovskii Bryansk State University 2017-03-01
• Series: Naučno-Tehničeskij Vestnik Brânskogo Gosudarstvennogo Universiteta
• Subjects: Data mining; scoring model; deductor; classification; logistic regression; Fine&Coarse Classing
• Online Access: http://ntv-brgu.ru/wp-content/arhiv/2017-N1/2017-01-08.pdf

The aim of the study is to build a scoring model based on Deductor analytical platform. Problems solved by this method are introduced to actualizing the research. The source data for customer response prediction is the scoring profile of the OTP Bank. Modeling process and Fine&Coarse Classing Deductor’s handler are described. Fine&Coarse Classing is capable of reducing the unique values and allows to analyze iteratively and to improve the quality and speed of model building. The weight of evidence, information value and the value of the objective function enable the fundamental parameters to be determined from experimental results. The conclusion is made that using Deductor have several advantages over manual calculations.