🧠 Loan Approval Prediction System

📌 Overview

This project implements an end-to-end machine learning pipeline to predict loan approval status based on applicant financial and demographic data. The system includes preprocessing, feature engineering, model training, evaluation, and prediction on unseen data.

---

⚙️ Technologies Used

• Python
• Pandas, NumPy
• Matplotlib
• Scikit-learn

---

Dataset

This project uses a dataset from Kaggle.

Note: Dataset is subject to its own license (CC0 / CC BY).

---

📂 Project Workflow (Notebook Explanation)

---

🔹 1. Import Libraries & Load Dataset

We import required libraries and load the dataset into a DataFrame. This allows us to perform data analysis and preprocessing. Initial rows are displayed to understand the structure.

import pandas as pd

df = pd.read_csv("loan.csv")
df.head()

---

🔹 2. Data Inspection

We inspect data types, structure, and missing values. This step helps identify issues such as null values and categorical variables. Understanding the dataset is crucial before preprocessing.

df.info()
df.isnull().sum()

---

🔹 3. Handling Missing Values

Missing values are handled using appropriate statistical methods. Median is used for numerical features to reduce outlier impact. Mode is used for categorical features to preserve distribution.

df['LoanAmount'] = df['LoanAmount'].fillna(df['LoanAmount'].median())
df['Loan_Amount_Term'] = df['Loan_Amount_Term'].fillna(df['Loan_Amount_Term'].median())

df['Gender'] = df['Gender'].fillna(df['Gender'].mode()[0])
df['Dependents'] = df['Dependents'].fillna(df['Dependents'].mode()[0])
df['Self_Employed'] = df['Self_Employed'].fillna(df['Self_Employed'].mode()[0])
df['Credit_History'] = df['Credit_History'].fillna(df['Credit_History'].mode()[0])

---

🔹 4. Feature Engineering

New features are created to improve model performance. These features capture financial relationships more effectively. They help the model learn better patterns.

df['Total_Income'] = df['ApplicantIncome'] + df['CoapplicantIncome']
df['EMI'] = df['LoanAmount'] / df['Loan_Amount_Term']
df['DTI'] = df['LoanAmount'] / df['Total_Income']

---

🔹 5. Dropping Irrelevant Features

The Loan_ID column is removed as it has no predictive value. Keeping irrelevant features can reduce model performance. This step ensures cleaner input data.

df = df.drop('Loan_ID', axis=1)

---

🔹 6. Target Encoding

The target variable is converted into numerical form. Machine learning models require numeric inputs. This allows classification algorithms to process the target.

df['Loan_Status'] = df['Loan_Status'].map({'Y': 1, 'N': 0})

---

🔹 7. Feature & Target Separation

The dataset is divided into input features and output labels. This separation is required for supervised learning. Models learn patterns from features to predict the target.

X = df.drop('Loan_Status', axis=1)
y = df['Loan_Status']

---

🔹 8. Categorical Encoding

Categorical variables are converted into numerical format. One-hot encoding creates binary columns for each category. This prevents models from misinterpreting categorical values.

X = pd.get_dummies(X, drop_first=True)

---

🔹 9. Train-Test Split

The dataset is split into training and testing sets. Training data is used to build the model. Testing data evaluates how well the model generalizes.

from sklearn.model_selection import train_test_split

X_train, X_test, y_train, y_test = train_test_split(
    X, y, test_size=0.2, random_state=42
)

---

🔹 10. Feature Scaling

Feature scaling standardizes the range of variables. This is important for models like SVM that depend on distance. It ensures all features contribute equally.

from sklearn.preprocessing import StandardScaler

scaler = StandardScaler()
X_train = scaler.fit_transform(X_train)
X_test = scaler.transform(X_test)

---

🔹 11. Model Training

Multiple models are trained to compare performance. Different algorithms capture patterns differently. This helps in selecting the best-performing model.

from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier
from sklearn.svm import SVC

lr = LogisticRegression()
rf = RandomForestClassifier()
svm = SVC()

lr.fit(X_train, y_train)
rf.fit(X_train, y_train)
svm.fit(X_train, y_train)

---

🔹 12. Model Evaluation

Models are evaluated using accuracy score. This helps compare performance across algorithms. The best model is selected based on results.

from sklearn.metrics import accuracy_score

print(accuracy_score(y_test, lr.predict(X_test)))
print(accuracy_score(y_test, rf.predict(X_test)))
print(accuracy_score(y_test, svm.predict(X_test)))

---

🔹 13. Confusion Matrix & Classification Report

Detailed evaluation metrics are computed. Precision, recall, and F1-score provide deeper insights. This helps understand model strengths and weaknesses.

from sklearn.metrics import confusion_matrix, classification_report

print(confusion_matrix(y_test, svm.predict(X_test)))
print(classification_report(y_test, svm.predict(X_test)))

---

🔹 14. ROC Curve

ROC curve evaluates model performance at different thresholds. It shows trade-off between true positive and false positive rates. AUC score indicates overall model quality.

from sklearn.metrics import roc_curve, auc

---

🔹 15. Hyperparameter Tuning

GridSearchCV is used to optimize model parameters. This improves model performance and generalization. It helps find the best combination of parameters.

from sklearn.model_selection import GridSearchCV

param_grid = {
    'n_estimators': [100, 200],
    'max_depth': [5, 10]
}

grid = GridSearchCV(RandomForestClassifier(), param_grid, cv=5)
grid.fit(X_train, y_train)

best_rf = grid.best_estimator_

---

🔹 16. Feature Importance (Explainable AI)

Feature importance identifies key influencing factors. This improves interpretability of the model. It helps understand decision-making.

import pandas as pd

importances = best_rf.feature_importances_
features = X.columns

pd.Series(importances, index=features).nlargest(10)

---

🔹 17. Cross-Validation

Cross-validation checks model stability across folds. It reduces risk of overfitting. Provides a more reliable performance estimate.

from sklearn.model_selection import cross_val_score

scores = cross_val_score(best_rf, X, y, cv=5)
print(scores.mean())

---

🔹 18. Test Dataset Preprocessing

The unseen dataset is processed similarly to training data. Consistency ensures correct model predictions. Feature alignment is crucial.

test_df = pd.get_dummies(test_df, drop_first=True)
test_df = test_df.reindex(columns=X.columns, fill_value=0)

---

🔹 19. Predictions on Unseen Data

The trained model is applied to new data. This simulates real-world deployment. Predictions are generated for each sample.

predictions = best_rf.predict(test_df)

---

🔹 20. Output Generation

Predictions are saved in CSV format. This allows submission or further analysis. Output follows required structure.

output.to_csv("submission.csv", index=False)

---

🔹 21. Final Results Display

Final accuracy is displayed on test data. Prediction distribution can also be analyzed. This summarizes overall performance.

print("Accuracy:", accuracy_score(y_test, best_rf.predict(X_test)))

---

📊 Key Insights

• SVM achieved the highest accuracy
• Random Forest provided better interpretability
• Credit history and income are key influencing factors
• Model performs better in predicting approvals than rejections

---

🎯 Conclusion

This project demonstrates a complete machine learning workflow including preprocessing, feature engineering, model comparison, tuning, and explainability. The system is capable of predicting loan approval and simulating real-world deployment.

---

🚀 Future Improvements

• Improve class imbalance handling
• Deploy using Streamlit
• Add real-time prediction system

---

License

Code is licensed under Apache 2.0. Dataset belongs to original source and is not redistributed.