| Abstract: | Due to the wide availability of computer, information and communication technologies data are being generated massively today, especially in financial institutions and banks data are being generated massively on regular basis. Microfinances are one of such institutions that collect, process and store huge amounts of records from time to time and therefore deal with large amount of data. On the other hand, Ethiopian Microfinances are facing problems in loan risk assessment and managing portfolio at risk. Currently Ethiopian microfinance institutions loan risk assessment and granting loan to the borrower’s is conducted in a traditional manner depends on the loan approval team views and believes, Moreover, such way of risk assessment creating inefficiency in quality of identifying borrower’s characteristics before granting the loan. If the microfinance institutions (MFIs) do not manage their loan risks well, they are likely to fail to meet their social and financial objectives. The existing past and historic data related to loan borrower and loan characteristics could be actionable and usable for loan risk assessment with the help of Machin learning algorithms. This study was conducted to demonstrate the practical methods, experiments and datasets with machine learning to assist Ethiopian MFIs through building a classification and prediction model which supports in prediction of a new loan borrower’s status (Active or Defaulter) when the loan decision making in the microfinance institutions. The classification and prediction model are built based on the MFIs loan borrowers’ data obtained from the selected seven (Aggar, Harbu,Vision Fund,Pease,Oromia,Nisir and wosasa) microfinance institutions in Ethiopia. Necessary preprocessing activities have been applied to clean and make it ready for the Experimentation. Then, the four algorithms used were SVM, KNN, Naïve Bayes and logistic regression. The RStudio with R programing was used to simulate all the experiments. Confusion matrix was used to calculate the accuracy, specificity, sensitivity and precision were used to evaluate the performance of the models and Cross table was used to visualize the performance of the models. The results of the experiment show high precision, so that the models can be used in detecting and predicting defaulter (risky) loan applicants. The KNN classifier produced an accuracy of 99.91%, the SVM classifier produced an accuracy of 92.4%, logistic regression model also produced an accuracy of 93.8%, and Naïve Bayes classifier produced an accuracy of 83.8 %. |
