💳 CodeAlpha Credit Scoring Model

A Deep Learning–powered Credit Scoring System with an interactive Gradio Web UI

---

📖 Project Overview

This project builds a Deep Learning Neural Network model to predict a customer's credit score category (Good / Standard / Poor) based on financial and behavioural attributes.

Credit scoring is used by banks and financial institutions to determine whether a customer is eligible for loans or credit cards. This model analyses customer financial history and predicts their creditworthiness, helping reduce financial risk.

---

🎯 Problem Statement

Build a machine learning pipeline that predicts whether a person is creditworthy based on their financial history and demographic details.

Why it matters:

• ✅ Reduce loan defaults
• ✅ Improve credit approval decisions
• ✅ Manage financial risk effectively
• ✅ Provide explainable, confidence-backed predictions

---

🧠 Model Architecture — Deep Neural Network (MLP)

The model uses scikit-learn's MLPClassifier with the following design:

Component	Details
Architecture	4 hidden layers: 256 → 128 → 64 → 32 neurons
Activation	ReLU (non-linearity)
Optimizer	Adam with adaptive learning rate
Regularisation	L2 (alpha = 0.001 → 0.005 after tuning)
Early Stopping	Yes — patience = 20 epochs
Batch Size	64
Validation Split	10% held out during training
Cross-Validation	2-fold Stratified K-Fold

---

🔄 Machine Learning Pipeline

Raw CSV Data
    │
    ▼
┌─────────────────┐
│  1. Load Data   │  ── train.csv & test.csv
└────────┬────────┘
         │
         ▼
┌─────────────────┐
│  2. EDA         │  ── Correlation Heatmap, Feature Distributions
└────────┬────────┘
         │
         ▼
┌──────────────────────┐
│  3. Preprocessing    │  ── Fill NaN, Encode, Scale (StandardScaler)
│     + SMOTE (opt.)   │  ── Handle class imbalance if available
└────────┬─────────────┘
         │
         ▼
┌──────────────────────┐
│  4. Train MLP        │  ── Deep Neural Network (256→128→64→32)
│     + Cross-Val      │  ── 2-Fold Stratified CV
└────────┬─────────────┘
         │
         ▼
┌──────────────────────┐
│  5. Hyperparameter   │  ── Refined alpha=0.005, lr=0.0005
│     Tuning           │
└────────┬─────────────┘
         │
         ▼
┌──────────────────────┐
│  6. Evaluation       │  ── Accuracy, ROC-AUC, Confusion Matrix
│     + Plots          │  ── Loss Curve, Feature Importance, ROC Curve
└────────┬─────────────┘
         │
         ▼
┌──────────────────────┐
│  7. Save Artifacts   │  ── best_model.pkl, scaler.pkl, metadata.pkl
└────────┬─────────────┘
         │
         ▼
┌──────────────────────┐
│  8. Predict          │  ── final_predictions.csv from test.csv
└──────────────────────┘

---

📊 Outputs & Visualisations

Output File	Description
correlation_heatmap.png	Heatmap showing feature correlations
feature_distributions.png	Distribution plots for top numeric features
feature_importance.png	Permutation importance bar chart (top 20)
training_loss_curve.png	Training loss + validation accuracy over epochs
roc_curve.png	ROC curve (binary classification)
final_predictions.csv	Predicted credit scores for test dataset

📋 Sample Predictions

ID	Customer_ID	Predicted_Score
0x160a	CUS_0xd40	Good
0x160b	CUS_0xd40	Standard
0x160c	CUS_0xd41	Poor

---

🖥️ Interactive Web UI (Gradio)

The project includes a Gradio-powered web dashboard (app.py) with:

• 🎛️ Input sliders and dropdowns for all financial features
• 🎯 Real-time credit score prediction (Good / Standard / Poor)
• 📊 Confidence chart — probability breakdown per class
• 📌 Feature importance chart — top 10 most influential features
• 🌙 Dark theme with gradient UI design

UI Input Fields

Section	Fields
👤 Personal Info	Name, Age, Monthly Income, Occupation
💳 Credit Behaviour	Credit Mix, Min Amount Paid, Delayed Payments, Days Past Due
📊 Financial Details	Outstanding Debt, Credit Utilization %, Interest Rate, Credit History Age, No. of Credit Inquiries

---

🗂️ Project Structure

CodeAlpha_Credit_Scoring_Model/
│
├── dataset/
│   ├── train.csv                   # Training data
│   └── test.csv                    # Test data
│
├── model/                          # Auto-generated after training
│   ├── best_model.pkl              # Trained MLP model
│   ├── scaler.pkl                  # StandardScaler
│   ├── label_encoders.pkl          # LabelEncoders for categorical features
│   └── metadata.pkl                # Feature names, accuracy, importances
│
├── credit_scoring_model.py         # Full deep learning training pipeline
├── app.py                          # Gradio web UI
├── requirements.txt                # Python dependencies
├── README.md                       # Project documentation
│
├── correlation_heatmap.png         # EDA output
├── feature_distributions.png       # EDA output
├── feature_importance.png          # Permutation importance chart
├── training_loss_curve.png         # Loss curve during training
├── roc_curve.png                   # ROC curve (binary)
└── final_predictions.csv           # Test set predictions

---

⚙️ Installation

1. Clone the Repository

git clone https://github.com/Venu200723/CodeAlpha_Credit_Scoring_Model.git
cd CodeAlpha_Credit_Scoring_Model

2. Install Dependencies

pip install -r requirements.txt

---

🚀 How to Run

Step 1 — Train the Model

python credit_scoring_model.py

This will:

• Load and preprocess the dataset
• Train the Deep Neural Network
• Save model artifacts to model/
• Generate all visualisation plots
• Output final_predictions.csv

Step 2 — Launch the Web UI

python app.py

Open your browser at: http://127.0.0.1:7860

---

🛠️ Technologies Used

Library	Purpose
Python 3.8+	Core language
Pandas	Data manipulation
NumPy	Numerical operations
Scikit-learn	MLPClassifier, preprocessing, evaluation
Imbalanced-learn	SMOTE for class balancing (optional)
Matplotlib	Plotting and visualisations
Seaborn	Statistical visualisations
Gradio	Interactive web UI
Joblib	Model serialisation

---

📈 Model Evaluation Metrics

• ✅ Accuracy Score
• ✅ ROC-AUC Score (macro-average for multi-class)
• ✅ Confusion Matrix
• ✅ Classification Report (Precision, Recall, F1)
• ✅ 2-Fold Stratified Cross-Validation
• ✅ Permutation Feature Importance

---

👨‍💻 Author

Venu Gopal R
B.Tech – Artificial Intelligence & Data Science

---

📄 License

This project is licensed under the MIT License — feel free to use, modify, and distribute.

---

_{Built with ❤️ as part of the CodeAlpha Internship Program}