obinna-Muonanu/African-Loan-Default-Prediction

# African-Loan-Default-Prediction ## Project Overview Financial institutions need to predict loan defaults to reduce risks and make better lending decisions. In Africa's growing financial markets, with different types of customers and changing economic conditions, it is very important to accurately assess the risk of default. The goal of this project is to develop a generalisable and robust machine learning model to predict the likelihood of loan defaults for both new and existing customers. This project is of great importance as it will help financial institutions make better decisions by predicting the likelihood of loan defaults. This can lead to: - Identifying high-risk loans early allows institutions to take preventive measures, reducing financial losses. - By managing risks effectively, institutions can confidently expand their operations in dynamic and growing markets, such as those in Africa. - Reducing defaults contributes to a more stable financial system, benefiting both lenders and borrowers. ## Data Handling The data used for this project was provided by a private asset manager that operates in several financial markets across Africa. You can access the data here: Data Link ### Data Preprocessing 1. **Loan Amount Adjustment**: - Set the minimum values for "Loan Amount" and "Loan Amount to Repay" to 1000. - This was based on findings during EDA, where some loans were unrealistically small (e.g., 2 KHS or 2 Ghana Cedis). 2. **Merging External Data**: - Added external data from the Federal Reserve Bank related to the countries of interest to improve the model's accuracy. 3. **Handling Missing Values**: - Filled in missing values to ensure the dataset was complete. 4. **Encoding Categorical Features**: - Used label encoding to convert categorical variables into numerical form. 5. **Feature Engineering**: - Created new features to add more information to the model: - **Loan to Interest Rate**: Total loan amount × country-specific lending interest rate ( …