Loan Default Prediction

Predicting loan default using machine learning models.
This project was developed as part of a data science challenge by Coursera Project Network .

---

📊 Project Overview

Loan default prediction is a critical task for financial institutions, helping them minimize losses and manage risk.
This project applies both traditional machine learning methods to predict loan defaults on a large, highly imbalanced dataset.

---

🛠️ Methods Used

• Exploratory Data Analysis (EDA):

  • Distribution checks
  • Class imbalance visualization
  • Correlation heatmaps

• Data Preprocessing:

  • Handling missing values
  • Encoding categorical features
  • Standard scaling numerical features
  • Train-test split with stratification

• Models Implemented:

  • Logistic Regression (balanced weights)
  • Random Forest
  • Gradient Boosting
  • LightGBM (with scale_pos_weight)

• Evaluation Metrics:

  • ROC-AUC Score (main metric)
  • Confusion Matrices

---

📈 Results

Model	ROC-AUC
Logistic Regression	0.69
Random Forest	0.68
Decision Tree	0.67
LightGBMt	0.76

• LightGBM achieved the best performance.
• Handling class imbalance (via class_weight and pos_weight) significantly improved ROC-AUC.

---

Install Dependencies

git clone https://github.com/b-tanyileke/deeplearn_pipeline_optimizer.git
pip install -r requirements.txt