AN EVALUATION OF CLASSIFICATION PERFORMANCE OF ARTIFICIAL NEURAL NETWORK AND LOGISTIC REGRESSION AS LOAN SCORING MODELS

Purpose: Application of credit risk evaluation techniques has continued to receive more research attention in the advanced economies than emerging economies. However, the findings on the classification performance of these techniques have been mixed. The current paper aims at assessing the classification performance of Artificial Neural Network (ANN) and Logistic Regression Model (LRM) in credit transaction using data from Nigeria emerging economy. Design/methodology/approach: 2,300 data samples comprising of fully and partially recovered loan accounts were obtained from the databases of eleven commercial banks and sixteen primary mortgage institutions practicing in Lagos metropolis, Nigeria. Also, data on 14 variables comprising one dependent variable (loan recovery status) and thirteen independent variables were collected on each of the data samples. To construct LRM & ANN models, the total samples were subdivided into training/validation and testing samples. 73% of the total samples (1,679) were used for training and validation of the models while 27% of the total samples (613) were used in testing the classification performance of LRM and ANN using overall accuracy, specificity, sensitivity, Type I and Type II errors as criteria for performance measurement. SPSS version 21 was adopted for data analysis. Findings: The result of the analysis reveals among others that LRM and ANN models generated good overall accuracy values of 76.6% and 91% respectively. However, the performance of ANN is comparatively more efficiently better than that of LRM in detecting ‘good’ loan applicants, ‘bad’ loan applicants and in generating lower Type I and Type II errors than LRM. The use of ANN is therefore recommended as a credit risk evaluation technique due to its consistent performance across the performance metrics. Research limitation/implications: The findings of the paper provide input for lending decision in lending institutions in Nigeria which is capable of minimizing bad debts & non-performing loans thereby enhances stability in financial institutions. However, the finding of the current research is of country-specific, further study may compare the classification capacity of LRM and ANN across advanced and emerging economies. Also, future studies may adopt larger samples than the ones adopted in the current study for more inclusive researches. Originality/value: As noted above, studies on classification performance of ANN & LRM have been well documented in the advanced economies like UK, USA, China etc, the current paper extends the frontiers of knowledge to the existing body of literature in the advanced economies