🧠 EasyVisa ML Predictive Modeling Project

📌 Overview

This project focuses on building a robust machine learning solution for EasyVisa, a visa services provider, to predict the outcome of U.S. visa applications (Certified or Denied). The model aims to optimize client screening, improve operational decision-making, and drive business insights.

---

🎯 Objective

To develop a predictive model using advanced machine learning techniques that:

• Classifies visa applications as Certified or Denied
• Identifies the most influential factors affecting certification
• Assists EasyVisa in prioritizing strong applications

---

📊 Data Summary

• Source: Internal records from EasyVisa
• Size: 25,480 visa applications
• Target Variable: case_status (Certified = 1, Denied = 0)
• Features: Education level, job experience, wage, employer size, employment region, full-time status, continent of origin, etc.

---

🔧 Methodology

🧹 Data Preprocessing

• No missing or duplicate values
• Fixed negative values and standardized wage units
• Handled outliers through capping or log transformation
• Feature engineering: One-hot encoding, binary flags, company age derivation

⚖️ Class Imbalance Handling

• Oversampling: SMOTE
• Undersampling: RandomUnderSampler

🧠 Model Building

• Algorithms Used:
  • Decision Tree
  • Random Forest
  • Bagging Classifier
  • AdaBoost
  • Gradient Boosting (GBM)
  • XGBoost
• Evaluation Metrics: Accuracy, Recall, Precision, F1 Score, ROC AUC
• Cross-validation: Stratified K-Fold

🔍 Hyperparameter Tuning

Used RandomizedSearchCV to optimize:

• n_estimators, learning_rate, max_depth, min_samples_leaf, etc.

---

🏆 Final Model: Gradient Boosting (GBM)

✅ Validation Performance

Metric	Score
Recall	0.8104
F1 Score	0.7968
Precision	0.7836
Accuracy	0.7239
ROC AUC	0.7239

Why GBM?

• Balanced performance across all metrics
• Excellent generalization
• High interpretability

---

🔍 Key Insights

• 🎓 Education: Higher degrees (Master's, Doctorate) improve approval odds
• 👨‍💼 Experience: Prior job experience is a strong predictor
• 🌍 Geography: South & Midwest regions yield better certification rates
• 💵 Wage: Higher and yearly wages are linked to approval
• 🧾 Full-Time Status: Full-time jobs are more likely to be approved

---

🚀 Business Recommendations

• Implement a pre-screening tool using GBM to flag high-risk applications
• Educate clients about factors that influence approval
• Focus resources on candidates with high approval likelihood
• Deploy a visa risk dashboard for internal triage
• Retrain model periodically to adapt to policy changes

---

📈 Future Enhancements

• Include job titles, industry, employer history, and application types
• Use application-level metadata (e.g., service center, benefits)
• Develop a dynamic risk scoring system to prioritize cases

---

🗂️ Project Structure

📦 easyvisa-ml-prediction/ ├── data/ ├── notebooks/ ├── models/ ├── src/ │ ├── preprocessing.py │ ├── modeling.py │ └── evaluation.py ├── reports/ ├── requirements.txt └── README.md

---

💡 Tech Stack

• Python (Pandas, Scikit-learn, XGBoost, Imbalanced-learn)
• Jupyter Notebooks
• Matplotlib / Seaborn
• Git / GitHub

---

🤝 Contributing

Contributions are welcome. Please fork the repo and open a pull request with clear documentation of changes.

---

📜 License

This project is licensed under the MIT License. See LICENSE for more details.