fmmido/ai-tax-customs-evasion-prevention

AI Platform for Detecting and Preventing Tax and Customs Evasion in Egypt: Anomaly Detection, NLP Analysis, and Real-Time Alerts. # AI Platform for Preventing Tax and Customs Evasion ## Overview This AI platform detects fraudulent activities in tax declarations and customs invoices using machine learning and NLP. Key features: - **Anomaly Detection**: Identifies unusual patterns in transaction amounts, HS codes, and values (e.g., under-valuation in imports). - **NLP Analysis**: Extracts entities from invoices (e.g., misclassified goods) using Hugging Face models. - **Real-Time Alerts**: Flags high-risk cases (e.g., evasion probability >70%) and generates reports. - **Self-Updating**: Retrains weekly on new data via GitHub Actions. Built for Egypt's tax/customs authorities, using simulated data from Kaggle (Egyptian VAT datasets) and public APIs. Aligns with digital tax reforms for Vision 2030. ## Features - Data Loading: Simulated invoices with features like amount, HS code, origin country. - Detection: Isolation Forest for anomalies + BERT for text classification (fraudulent descriptions). - Alerts: JSON reports with risk scores and mitigation suggestions. - Dashboard: Streamlit UI for querying and visualizing flagged cases (run `streamlit run dashboard.py`). ## Installation and Running 1. Clone the repo: `git clone github.com` 2. Install requirements: `pip install -r requirements.txt` 3. Load data: `python data_loader.py` 4. Train detector: `python evasion_detector.py` 5. Generate alerts: `python alert_generator.py` 6. Run dashboard: `streamlit run dashboard.py` ## Data Sources - Kaggle Egyptian Tax Dataset: [kaggle.com]. - HS Code API: UN Comtrade for customs validation. - Real-Time: Integrate with Egypt Tax Authority APIs (mocked here). ## Technologies - Python 3.10+ - scikit-learn for anomaly detection. - Transformers (Hugging Face) for NLP. - Pandas for data handling. - Streamlit for UI. - GitHub Actions for automation.