| Field | Details |
|---|---|
| Project Name | US Visa Approval Prediction |
| Prepared By | Md. Alahi Almin Tansen |
| Revision Number | 1.0 |
| Last Revision Date | 02-09-2025 |
| Revision | Date | Changes Made | Author |
|---|---|---|---|
| 1.0 | 02-09-2025 | Initial Draft | Md. Alahi Almin Tansen |
- The system predicts whether a US visa application will be approved or rejected based on applicant & employment features.
- End-users → Visa applicants, consultants, researchers.
- Interaction → Users input data via UI/API → Model predicts result → Response displayed.
flowchart LR
U[User] -->|Input Features| UI[Web / API UI]
UI --> API[Backend API Service]
API --> ML[ML Model Service]
ML --> DB[(Database: Visa Data)]
ML --> API
API --> U[Prediction Result]
subgraph Deployment
API
ML
DB
end
- Web-based input form (React, Django template, or Streamlit)
- REST API endpoints for external integrations
- Fields: continent, education, experience, wage, etc.
-
API Layer:
- Expose endpoints:
/predict,/train,/health - Framework: FastAPI / Django REST Framework
- Expose endpoints:
-
Business Logic:
- Preprocess user input → feature encoding/scaling
- Call trained ML model for prediction
- Model Type: Classification (KNN)
- Input: Applicant features
- Output: Approval Probability + Final Decision (Yes/No)
- Model Training Pipeline:
- Data ingestion → Preprocessing → Training → Validation → Model Registry
- Historical visa dataset (structured CSV/Mongo DB)
- Store: training data, feature metadata, model versions, logs
- Docker: Containerize ML + API
- Cloud Services: AWS/GCP/Azure (compute + storage)
- CI/CD: GitHub Actions with self-hosted runner
- Monitoring: API health, model drift tracking
sequenceDiagram
participant User
participant UI
participant API
participant ML
participant DB
User->>UI: Submit visa application features
UI->>API: Send features as JSON
API->>ML: Preprocess & Predict
ML->>DB: Fetch metadata / feature mapping
ML->>API: Return Prediction (Approved/Rejected)
API->>UI: Display result to User
- Scalability → Handle multiple predictions concurrently
- Performance → Response < 2 seconds
- Security → Input validation, secure APIs
- Reliability → Deployed with redundancy
- Maintainability → Modular services, CI/CD pipelines
- Programming: Python
- Frameworks: FastAPI / Django REST Framework
- ML: scikit-learn, KNN
- Data Storage: MongoDB / CSV files
- Deployment: Docker, GitHub Actions, Cloud Hosting
- Monitoring: Prometheus + Grafana, Evidently AI