An ensemble model of outlier detection with random tree data classification for financial credit scoring prediction system

Abstract

Recently, Financial Credit Scoring (FCS) becomes an essential process in the finance industry for assessing the creditworth of individual or financial firms. Several artificial intelligence (AI) models have been already presented for the classification of financial data. However, the credit as well as financial data generally comprises unwanted and repetitive features which lead to inefficient classification performance. To overcome this issue, in this paper, a new financial credit scoring (FCS) prediction model is developed by incorporating the process of outlier detection (OD) process (i.e. misclassified instance removal) prior to data classification. The presented FCS model involves two main phases namely misclassified instance removal using Naïve Bayes (NB) Tree and Random Tree (RT) based data classification. The presented NB-RT model is validated using the Benchmark German Credit dataset under different validation parameters. The extensive experiments exhibited that a maximum classification accuracy of 90.3% has been achieved by the proposed NB-RT model.