Leighndeigh/AnalystLab-Africa-Week-2---Loan-Prediction

This project prepares the Kaggle Loan Prediction Dataset for machine learning. The goal is to predict whether a loan application is approved (Loan_Status). ## AnalystLab Africa – Week 2: Loan Prediction Feature Engineering ## Project Overview This project prepares the Kaggle Loan Prediction Dataset for machine learning. The goal is to predict whether a loan application is approved (Loan_Status). Target: Loan_Status Approved: 422 (68.7%) Rejected: 192 (31.3%) ## Feature Engineering TotalIncome = ApplicantIncome + CoapplicantIncome LoanToIncome = LoanAmount / TotalIncome Loan_ID removed because it is an identifier. ## Missing Values Categorical fields were imputed with the mode. Numerical fields were imputed with the median because the dataset contains extreme income and loan observations. ## Encoding Loan_Status: binary label encoding, N=0 and Y=1. Dependents: ordinal encoding from 0 to 3. Nominal categorical variables: one-hot encoding with drop_first=True. ## Scaling RobustScaler was used on continuous numerical features because it uses the median and IQR and is less sensitive to extreme observations. ## Outliers The IQR method was used. Monetary and ratio features were capped at IQR bounds rather than deleting records. Binary Credit_History and discrete Loan_Amount_Term were not treated as ordinary continuous outliers. ## Feature Selection Correlation analysis was used to identify redundancy. Random Forest feature importance was used as an exploratory model-based ranking. The strongest features included: - Credit_History - LoanToIncome - TotalIncome - ApplicantIncome - LoanAmount ## Key Finding Credit history is the strongest individual feature in the Random Forest importance analysis, while income and loan-affordability measures are also prominent. The dataset contains no duplicate records and all missing values were resolved. ## Technologies Python, Pandas, NumPy, Matplotlib, Seaborn, Scikit-learn, Jupyter Notebook.