Loan Classification Project

Overview

This project involves predicting loan approval status using machine learning techniques. The dataset contains information about applicants, loan details, and credit history, providing a comprehensive foundation for exploring and modeling loan approval patterns. Advanced techniques like hyperparameter tuning, feature analysis, and ensemble learning were used to achieve optimal results.

---

Features

Key Contributions

• End-to-End Pipeline:
  • Data cleaning, preprocessing, and exploration.
  • Implementation of multiple machine learning models.
  • Detailed performance analysis and visualization.
• Advanced Techniques:
  • Randomized hyperparameter tuning using RandomizedSearchCV.
  • Feature importance analysis to explain model predictions.
  • Comparative analysis across different algorithms.
• Interpretability:
  • Visualizations highlighting data trends and model performance.
  • Insights into factors influencing loan approvals.

---

Dataset Overview

Columns in the Dataset

• Applicant Demographics:
  • person_age, person_gender, person_education, person_income.
• Loan Details:
  • loan_amnt, loan_int_rate, loan_percent_income, loan_intent.
• Credit History:
  • cb_person_cred_hist_length, credit_score, previous_loan_defaults_on_file.
• Target Variable:
  • loan_status (Approved/Rejected).

Dataset Size

• Rows: 45,000+
• Columns: 14 (After cleaning the data using Pandas).

---

Technologies Used

• Programming Language: Python
• Libraries:
  • Data Manipulation: pandas, numpy
  • Visualization: matplotlib, seaborn
  • Machine Learning: scikit-learn
  • Model Tuning: RandomizedSearchCV

---

Machine Learning Models

1. Decision Tree Classifier

• Methodology:
  • Tuned hyperparameters like tree depth, leaf size, and splitting criteria.
  • Feature importance analysis for interpretability.
• Performance:
  • Achieved 91.69% accuracy with high recall for approved loans.

2. AdaBoost Classifier

• Methodology:
  • Boosted weak learners using optimized Decision Tree base estimators.
  • Tuned parameters: learning rate, number of estimators, and base depth.
• Performance:
  • Achieved 93.12% accuracy, improving overall recall for rejected loans.

3. Support Vector Machine (SVM)

• Methodology:
  • Scaled features for compatibility with SVM.
  • Optimized C, gamma, and kernel type using randomized search.
• Performance:
  • Balanced accuracy with reduced overfitting.

---

Workflow

1. Data Exploration:

   • Analyzed relationships between applicant income, intent, and loan status.
   • Explored how demographic and credit history features affect approvals.

   

   

2. Data Preprocessing:

   • Handled missing data for critical features.
   • Encoded categorical variables into numerical format.
   • Log-transformed skewed features for stability.

3. Model Implementation:
   • Built and tuned Decision Tree, AdaBoost, and SVM models.
   • Conducted 5-fold cross-validation for robust evaluation.
4. Feature Importance Analysis:
   • Identified top predictors like income, credit score, and loan intent.
5. Evaluation:
   • Compared accuracy, precision, recall, and F1-scores across models.

---

Results

Model	Accuracy	Precision	Recall	F1-Score
Decision Tree	91.69%	94%	96%	95%
AdaBoost	93.12%	94%	97%	96%
SVM	TBD	TBD	TBD	TBD

---

Visualizations

Key Insights

• Feature Importance:
  • Decision Tree and AdaBoost identified key factors like income and credit history as the most significant predictors of loan approval.
• Loan Status Analysis:
  • Visualized distribution of income and loan intent by approval status.
• Model Performance:
  • Plotted precision-recall and ROC curves to assess model reliability.

---