A Non-Parametric-Based Computationally Efficient Approach for Credit Scoring

EasyChair Preprint 1568

Abstract

This research aimed at the case of credit scoring in risk management and presented the novel method for credit scoring to be used for default prediction. This study uses Kruskal-Wallis non-parametric statistic to form a computationally efficient credit-scoring model based on artificial neural network to study the impact on modelling performance. The findings show that new credit scoring methodology represents reasonable coefficient of determination and low false negative rate. It is computationally less expensive with high accuracy (AUC=0.99). Because of the recent respective of continued credit/behavior scoring, our study suggests to use this credit score for non-traditional data sources such as mobile phone data to study and reveal changes of client’s behavior during the time. This is the first study that develops a non-parametric credit scoring, which is able to reselect effective features for continued credit evaluation and weighted out by their level of contribution with a good diagnostic ability.

Keyphrases: Credit Scoring, Kruskal_Wallis statistic, Loan Default, neural network, non-traditional data

@booklet{EasyChair:1568,
  author    = {Afshin Ashofteh and Jorge Bravo},
  title     = {A Non-Parametric-Based Computationally Efficient Approach for Credit Scoring},
  howpublished = {EasyChair Preprint 1568},
  year      = {EasyChair, 2019}}