mplement an Efficient Loan Credibility Prediction Model for Tsedey Bank using a Data Mining Model

This study examines the effects of data mining on predicting the likelihood of loan defaults in Tsedey Bank, an Ethiopian bank. The research addresses the perennial problem of defaulted loans which pose risk to both lending institutions and the economy at large. The study was based on a sample of 3,000 loan portfolios and followed a systematic approach within the CRISP-DM framework. Phases of the methodology were data collection; data preprocessing; EDA; model creation; and model evaluation. The objective of the study was to assign loans at the bank to categories of Pass, Special Mention, Doubtful, Loss, and Substandard according to borrower characteristics, repayment behavior, and financial ratios. Multiple models were created and assessed to compare the results including Random Forest, Decision Trees, K-Nearest Neighbors (KNN) & Logistic Regression. These results also showed that the Decision Tree Classifier achieved the best accuracy with 99.83% compared to the other models. He also reached a high level of explain -ability. However, we could see a few problems with Logistic Regression concerning the minority class which is discussed in the next step. These results underscored both the necessity for improved predictive modeling and the necessity of resolving the class imbalance problem. The author concludes: “In light of the findings, the financial organizations should consider deploying ensemble classifiers such as Random Forest and Decision Trees based algorithms also take into account controlling class imbalances in datasets, and training of the model to reflect the up-to-date state of the market.” In that regard, this thesis highlights the importance of credit risk management and the usage of data mining techniques for a more transparent lending process and further economic growth in the area.