This repository contains the implementation of a loan default prediction model using XGBoost. The model is trained to predict whether a loan applicant is likely to default based on various features such as income, credit score, loan amount, etc.

The dataset used for training and evaluation contains information on loan applicants, including their financial profiles, employment details, and loan terms. It consists of both numerical and categorical features.

The project follows a systematic workflow, including:

1. Exploratory Data Analysis (EDA): Analyzing the dataset to understand the distributions and relationships of features.
2. Feature Engineering: Creating new features or transforming existing ones to improve model performance.
3. Data Preprocessing: Handling missing values, encoding categorical variables, and scaling numerical features.
4. Handling Imbalanced Data: Using techniques such as SMOTE to address class imbalance.
5. Model Selection and Hyperparameter Tuning: Experimenting with various classifiers and optimizing hyperparameters using techniques like GridSearchCV , StratifiedKFold.
6. Model Evaluation: Assessing model performance using metrics such as accuracy, F1-score, precision, recall, and AUC-ROC curve.
7. Selection of Best Model: Identifying the XGBoost classifier as the best-performing model based on evaluation results.

• Accuracy: 86.14%
• F1-score (Class 1): 83.91%
• Precision (Class 1): 98.3%
• Recall (Class 1): 73.91%
• AUC (Class 1): 91.8%

The XGBoost model demonstrates superior performance in predicting loan defaulters, achieving an accuracy of 86% and a high recall rate of 74%. This indicates that the model effectively identifies instances of defaulters while maintaining a reasonable precision score.

Potential areas for further improvement include:

• Experimenting with additional feature engineering techniques.
• Exploring advanced algorithms or ensemble methods.
• Conducting more extensive hyperparameter tuning to fine-tune model performance.
• Evaluating model robustness using cross-validation or validation on external datasets.