VisaIntel AI

Inspiration

Visa applications are often opaque, leaving applicants unsure why they are rejected. We wanted to create a system that predicts visa rejection risk transparently, explains why, and provides actionable steps to improve approval chances.

What it does

VisaIntel AI: Predicts visa rejection risk using an AI model Explains key factors influencing the decision with SHAP explainability Provides personalized improvement recommendations Displays results in a professional, fintech-style Streamlit UI

How I built it

Data: Synthetic, ethical dataset simulating real-world visa applications ML Model: Logistic Regression with balanced class weights Preprocessing: StandardScaler for numerical + OneHotEncoder for categorical features Explainable AI: SHAP for feature importance and human-readable insights UI: Streamlit app with tabs for Risk Assessment, Recommendations, and Model Transparency Evaluation: Separate evaluation.py for comprehensive model metrics, ROC-AUC, recall, confusion matrix, and technical reporting

Challenges I ran into

Handling imbalanced data (more approvals than rejections) Translating technical SHAP outputs into human-friendly explanations Designing a UI that balances professional aesthetics with clarity

Accomplishments that I'm proud of

Achieved a robust predictive model with strong recall (0.73) and ROC-AUC (0.78) Developed an end-to-end explainable AI pipeline Built a clean, user-friendly interface separating product experience and technical evaluation Created actionable, personalized recommendations for users

What I learned

The importance of explainability in AI for user trust How to handle imbalanced datasets and measure model performance beyond accuracy Building separation of concerns: product UI vs technical evaluation

What's next for VisaIntel AI

Add more visa types and country-specific rules Integrate real-world data (with privacy compliance) Enhance recommendation engine using advanced NLP Deploy as a cloud-accessible tool for hackathon judges and users

Built With

• matplotlib
• numpy
• pandas
• python
• scikit-learn
• seaborn-tools:-jupyter-notebook
• shap