Abstract:
The new evolving technologies bring new insights of supporting decision making in
different areas like finance, marketing, production, social world, healthcare and are
showing positive results. Devices and sensors become smart and are helping in a variety
of areas like predicting failure and preventive maintenance. The main risk for many
organizations comes from the low level of risk prediction. Banks and Micro Finance
Institutions are highly challenged by the low level of credit default risk estimation.
Loan status conditions in banks and microfinance institutions fall in the three classes of
normal loan, substandard loan and loss loan depending on the loan repayment of the
customer. This research aimed to develop a credit default prediction model by using loan
and customer information with data mining methods. To build a model a six step cross
industry standard for data mining process model was applied. For model building, python
data miner was used.
For this research we used 73,596 loan data from Amhara credit and saving institution. For
model building random forest, k-nearest neighbor and decision tree algorithms were
tested and evaluated. Random forest algorithm achieved the best accuracy of 96.81%.
The results of these algorithms were tested with accuracy measurement methods of
confusion matrix.