Research Experience 📚
Publications 📚
1. Portfolio Optimization with Stochastic Return Functions: An Algorithmic Approach
- Research Advisor: Dr. Roy E Welsch
- Status: Ongoing
- Lab: Laboratory for Information & Decision Systems, Massachusetts Institute of Technology (MIT)
2. Revisiting the Equity Premium Puzzle: Time Series Forecasting 1990–2012
- Status: Ongoing
- Independent Research: Department of Computer Science, Harvard University
3. Algorithmic Trading: Comparative Analysis of Quadrant Strategies
- Status: Ongoing
- Independent Research: Department of Computer Science & Statistics, Harvard University
4. Comparative Analysis of Time-Series Regression Techniques
- Independent Research: Department of Computer Science & Statistics, Harvard University
- Status: January 2025 [Under Publication Review]
- Details:
- Analyzed Ridge, Lasso, Weighted Norm, and Quantile Regression on financial time-series data using MSE, MAE, and R^2.
- Demonstrated Ridge’s predictive accuracy and Weighted Norm Regression’s robustness to outliers.
5. Enhancing Financial Factor Analysis with IPCA and Procrustes Alignment
- Research Advisor: Dr. Alexander Young – Department of Statistics, Harvard University
- Status: December 2024 [Under Publication Review]
- Details:
- Applied IPCA to estimate latent factors in high-dimensional financial data, leveraging observable characteristics.
- Enhanced factor interpretability with Procrustes alignment, reducing cross-validation discrepancies by 70%.
6. Cost Efficient Stock Using Forecasting with Enhanced LightGBM and Optuna
- Achievement: IEEE International Conference MoSICom
- Advisor: Dr. Tamizharasan PS (Ph.D. – CSE, National Institute of Technology, Tiruchirappalli)
- Publication: Click Here
- Optimized LightGBM model using Optuna, achieving a 15.2% annualized return and a 3.24 Sharpe ratio, significantly outperforming benchmark returns.
- Developed cost-awareness strategy to reduce false-positive errors, enhancing prediction reliability and lowering investment costs.
7. Dynamic Beta Variability in Foreign Exchange Returns Using Instrumented PCA
- Achievement: 2nd Place, National Undergraduate Research Competition, 2022 - Peer Reviewed By Abu Dhabi University.
- Advisor: Dr. Tamizharasan PS (Ph.D. – CSE, National Institute of Technology, Tiruchirappalli)
- Publication: Click Here
- Applied IPCA to build a flexible factor model, reducing FX data dimensionality and accommodating time-varying betas for superior out-of-sample predictability.
- Demonstrated economic significance by showing IPCA-based trading strategies outperformed PCA by 8%, especially for the Swiss Franc and Australian Dollar.
8. Deep Learning-Based Smart Parking Management System and Business Model
- Achievement: Published in Springer Conference - CVIP 2020, Singapore
- Advisor: Dr. Tamizharasan PS (Ph.D. – CSE, National Institute of Technology, Tiruchirappalli)
- Publication: Click Here
- Architected the workflow of ensemble techniques for detecting and classifying parking occupancy with 95% precision.
- Used TensorFlow for training and evaluation, improving F1 score, recall, and precision metrics.
9. Credit Risk Assessment Model for UAE’s Commercial Banks: A Machine Learning Approach
- Achievement: 2nd Place in Undergraduate Research Competition, 2021 - Peer Reviewed By Abu Dhabi University.
- Advisor: Dr. Parizad Dungore (Dubai Business School, University of Dubai)
- Publication: Click Here
- Developed a ML based credit risk model using Linear Discriminant Analysis and Adaboost, achieving 95.2% accuracy.
- Implemented and tested models like Logistic Regression and Decision Trees on 7M+ records, identifying key risk factors through feature selection.
10. Lithium-Ion Battery Life Prediction Based on Initial Stage-Cycles Using Machine Learning
- Achievement: Granted Intellectual Property Rights
- Advisor: Dr. Vilas Haridas Gaidhane (PhD - Delhi University)
- Publication: Click Here
- Developed a Gradient Boosting Trees model to predict lithium-ion battery life using initial 50-cycle charge/discharge data.
- Applied Kernel PCA to project data into higher-dimensional space, enhancing model robustness and prediction accuracy.
11. Real-Time Drowsiness Detection Framework Using Computer Vision to Prevent Car & Road Accidents
- Achievement: Granted Intellectual Property Right
- Advisor: Dr. Raja Muthalagu
- Publication: Click Here
- Developed and implemented a real-time drowsiness detection system using OpenCV’s Haar Cascade Classifier, achieving an accurate detection rate of drowsiness and distraction.
- Utilized Raspberry Pi 4+ and NoIR-V2 Pi camera for hardware implementation, ensuring efficient real-time processing and low energy consumption.
For further inquiries or collaborations, feel free to reach out to me at adityasaxena@g.harvard.edu ✉️