** Data Scientist | Quantitative Analyst | Mathematical Engineer**
Production-ready implementations of statistical models, machine learning systems, and quantitative trading strategies. Focus on high-performance computing, rigorous validation, and institutional-grade code quality.
High-Frequency Trading with Hawkes Processes
Production-grade multivariate Hawkes process implementation for order flow prediction.
| Metric | Result | Assessment |
|---|---|---|
| Sharpe Ratio | 86.98 | Exceptional |
| Win Rate | 62.5% | Strong |
| Speedup | 10,000x | UltraFast MLE |
- Key Innovations: O(N) recursive MLE (0.5s vs >1h), real-time production trading, comprehensive statistical validation
- Technologies: Numba, Cython, Time-series CV, Bootstrap inference
- Grade: A+ | Production Ready
Physics-Informed Neural Networks for Differential Equations
PINN implementation for solving forward and inverse PDE problems with energy conservation guarantees.
- Double pendulum system with domain decomposition (MSE < 1e-3)
- NLLSQ and VarPro methods for inverse problems
- GPU memory optimization and adaptive activation functions
- Residual network architecture with energy conservation losses
XGBoost-Based Claims Prediction System
Automated actuarial loss prediction with advanced feature engineering and text processing.
- XGBoost with hyperparameter optimization and GPU acceleration
- NLP processing for claim descriptions
- Segment-wise analysis and pricing optimization
- End-to-end ML pipeline (RMSE: 23,669)
LSTM-Based Risk Assessment Platform
Market risk evaluation using deep learning and statistical arbitrage detection.
- LSTM networks for market crash prediction
- Robust preprocessing pipeline (noise reduction: 8.7%)
- Trading simulation with risk metrics (Sharpe, Drawdown)
- Early stopping convergence in 10 epochs
Student Dropout Early Warning System
Machine learning approach for early detection of at-risk students using LightGBM.
- LightGBM with GPU acceleration
- Comprehensive feature engineering (socioeconomic factors)
- 88% accuracy in dropout prediction
- Interactive performance dashboards
- Statistical Modeling: Hawkes processes, Time-series analysis, Survival analysis
- Machine Learning: Gradient boosting, Neural networks, Bayesian methods
- Quantitative Finance: High-frequency trading, Risk management, Market microstructure
- Scientific Computing: PDE solvers, Numerical optimization, High-performance computing
| Category | Tools |
|---|---|
| Languages | Python, R, MATLAB, SQL |
| ML/DL | PyTorch, TensorFlow, scikit-learn, XGBoost, LightGBM |
| Scientific | NumPy, SciPy, Pandas, Numba, Cython, FEniCS |
| Data | PostgreSQL, Neo4j, MongoDB |
| MLOps | MLflow, Docker, GitHub Actions |
| Visualization | Matplotlib, Seaborn, Plotly, Jupyter |
ds-projects/
├── hawkes-order-flow/ # HFT with Hawkes processes (NEW)
├── neural-pdes-solver/ # Physics-informed neural networks
├── actuarial-loss-prediction/ # XGBoost claims prediction
├── market-risk-analysis/ # LSTM risk assessment
├── academic-performance/ # Student dropout prediction
├── pdes-simulations/ # FEM numerical solutions
└── README.md
- A biharmonic equation with discontinuous nonlinearities Eduardo Arias, Marco Calahorrano, Alfonso Castro Electronic Journal of Differential Equations, 2024
- LinkedIn: linkedin.com/in/eduardo-arias-3e0
- Email: mat.eduardo.arias@outlook.com
- Location: Ecuador
MIT License - See LICENSE for details.
All projects include comprehensive documentation, statistical validation, and production-ready code.