Abstract:
Credit score prediction is the most effective way of analyzing whether a potential client is
eligible for loan or not especially in financial institutions where class imbalance problems are prevalent. However, limited number of credit score prediction models in banking institutions take into consideration imbalance data and again, the best resampling technique to be applied with imbalanced data is still a challenge.
Therefore, in an attempt to address these problems, this research presents an empirical
comparison of various combinations of data imbalance resampling techniques and machine learning algorithms used to address this challenge of imbalance data. This study utilized credit score secondary data from bank of Kigali with 58096 customer transaction with 31 variables.
The time scope of data was limited to 2018-2019. Modelling the data and handling class
imbalance was done using python jupyter notebook libraries.
The credit score prediction from each combination were evaluated with F_Beta, F1_score,
precision, recall score to avoid biasness towards majority class hence taking into consideration effectiveness in each technique and model used which have not been considered in similar studies.
An experimental result was done in this research using resampling techniques and machine learning algorithms to enhance credit score prediction for bank of Kigali clients.
The findings suggest that combining oversampling technique and random forest algorithm yield the best prediction of 96.42% and F_Beta of 97.56% among the rest hence the most effective way of evaluating the eligibility of customers in the banking institution.
