Donia Besher
|
Exploring the intersection of deep learning, probabilistic forecasting, and spatiotemporal modelling.
About Me
I am Donia Besher, an MS/PhD student in Statistics at KAUST, supervised by Prof. Raphaël Huser, and a Sorbonne Mathematics graduate. Before joining KAUST, I worked as a Research Assistant at Sorbonne under the supervision of Prof. Tanujit Chakraborty, where I focused on developing forecasting frameworks. My research interests lie at the intersection of deep learning, probabilistic forecasting, and spatiotemporal modelling.
My journey began with a curiosity for mathematics and problem solving, which grew into a passion for uncovering patterns in complex data. Through my research, I have worked on forecasting exchange rates, modelling climate policy uncertainty, and developing probabilistic forecasting frameworks for temporal and spatiotemporal systems.
One of my favorite moments was watching my ranking change with every Kaggle submission and eventually reaching the top after countless experiments and tweaks. It reminded me that progress comes from curiosity, persistence, and being willing to learn from unexpected results.
Looking ahead, I am excited to explore methods for modelling complex spatiotemporal processes and extreme events under uncertainty.
Areas of Interest
Publications
Package Downloads
Kaggle Wins
GPA
Languages
Education
M.S./Ph.D. in Statistics (Data Science Track)
King Abdullah University of Science and Technology (KAUST)
Thuwal, Saudi Arabia
Selected Coursework
Bachelor's Degree in Mathematics
Sorbonne Université
Paris, France
Specialization Modules
Work Experience
Research Assistant
Sorbonne University, Center for Applied Mathematics, Statistics & Data Science
Abu Dhabi, UAE
Conducted research with Prof. Tanujit Chakraborty on jointly modelling nonlinearity and long memory in time series, a notoriously challenging combination, while advancing probabilistic and spatiotemporal forecasting through deep generative and neural long-memory models released as open-source software.
Peer Tutor
Sorbonne University
Abu Dhabi, UAE
Mentored second-year (L2) students in data science and statistics, fostering a collaborative learning environment and delivering tailored one-on-one sessions that led to significant academic improvement and increased confidence.
Science Communicator
“The Simplicity of Complexity”, Luxembourg Pavilion, Expo 2020
Dubai, UAE
Facilitated interactive board games based on open mathematical problems designed by the University of Luxembourg and addressed inquiries from the public regarding mathematical concepts.
Projects
Deep Generative Transformers for Probabilistic Spatiotemporal Forecasting
May 2024 – Aug 2026
Developed GEnformer, a lightweight, deep generative Transformer framework that integrates graph convolutions, stochastic noise injection, and Energy Score optimization to capture complex spatiotemporal dependencies for probabilistic forecasting. Released as the Genformer Python package, a modular PyTorch framework supporting training, inference, and uncertainty-aware prediction, implementing both the Enformer and Graph-Enformer (GEnformer) architectures.
Exchange Rate Forecasting
May 2024 – Aug 2026
Developed the NARFIMA (Neural AutoRegressive Fractionally Integrated Moving Average) model to forecast exchange rates of BRIC economies. NARFIMA integrates neural networks with fractional differencing to capture nonlinear patterns and long-term dependencies in financial time series. The framework was benchmarked against 16 baseline models with robustness and statistical-significance tests, and includes causal inference analyses, ablation studies justifying ARFIMA residual integration, and uncertainty quantification via conformal and simulation-based prediction intervals.
Probabilistic Forecasting of Climate Policy Uncertainty
2024 – 2026
Investigated drivers of the US Climate Policy Uncertainty (CPU) index using macroeconomic, financial cycle variables, and real-time public sentiment data to improve forecasting. This is the first framework for forecasting CPU with explicit emphasis on policy-relevant interpretability and uncertainty quantification, integrating Google Search indicators to capture early shifts in public attention and connecting forecasts to SDG 13 (Climate Action) and SDG 8 (Decent Work and Economic Growth).
Will They Get the Job?
Apr 2025 – May 2025
Developed the winning Kaggle competition machine learning model to predict student job placement outcomes from a highly imbalanced dataset with limited samples, nonlinear separability, and multicollinearity. The framework tackles these constraints by combining threshold-shifting and resampling techniques with a custom asymmetric kernel and robust ensemble learning.
Credit Card Debt Prediction
Mar 2024 – Apr 2024
Developed the winning Kaggle competition ensemble regression model for predicting credit card debt based on socio-economic factors dataset with outliers, nonlinearity, multicollinearity, and nonnormality. The framework addresses these challenges by combining robust loss optimization and nonlinear transformations with regularization techniques and ensemble regression.
Analysis of COVID-19 Mortality Data
Mar 2023 – May 2023
Does COVID-19 mortality data follow an exponential distribution? A probabilistic analysis was conducted to identify the underlying distribution and provide insights for pandemic management strategies. The project covers parameter estimation, numerical optimization, and goodness-of-fit testing to determine the best-fitting distribution. To see what the data revealed, check out the links below.
Publications & Open Source Packages
Publications
Neural ARFIMA for Forecasting BRIC Exchange Rates with Long Memory
Besher, D., Panja, M., Sengupta, S., Chakraborty, T. — Submitted to Studies in Nonlinear Dynamics & Econometrics (2026+)
Read the preprintProbabilistic Forecasting of Climate Policy Uncertainty: The Role of Macro-financial Variables and Google Search Data
Besher, D., Sengupta, A., Chakraborty, T. — In revision at Journal of Environmental Management (2026+)
Read the preprintOpen Source Packages
Genformer
PyPI · Python · DeveloperPyTorch-based Python package for lightweight, deep generative Transformer-based probabilistic forecasting, implementing the Enformer and GEnformer architectures for multivariate time series and spatiotemporal data.
Read the Documentation narfima
CRAN · R · Developer & MaintainerR package for forecasting time series exhibiting both nonlinear dynamics and long memory behavior, with support for exogenous variables, preprocessing, uncertainty quantification, and benchmarking.
Outreach & Engagement
Women in STEM Speaker
CCI France UAE
Nov 2024
Delivered a talk on the role of time series forecasting and AI in advancing the Sustainable Development Goals (SDGs), focusing on monetary policy for SDG 8 (Decent Work and Economic Growth) and climate challenges for SDG 13 (Climate Action).
Future of Energy Hackathon
TotalEnergies, Institut Français & La Fresque du Climat
Jun 2022
Investigated the causes and implications of climate change, designed sustainable development strategies for École 42, contributed to innovative solutions such as a self-generating energy keyboard, and presented proposals to a panel of experts.
Skills
Programming
Libraries & Tools
Data Science
Soft Skills
Achievements
Highest Honors & Kaggle Champion
Graduated with Mention Très Bien (GPA 4.0) from Sorbonne Université and ranked 1st in multiple Kaggle machine-learning competitions
Kaggle Certificate of Excellence
Ranked 1st in Kaggle machine-learning competitions (2024, 2025).
75% Merit-Based Scholarship
Awarded a 75% merit-based scholarship by Sorbonne University (2025).
50% Merit-Based Scholarship
Awarded a 50% merit-based scholarship by Sorbonne University (2023, 2024).
Ardian Foundation Scholarship
Scholarship for outstanding academic potential from Sorbonne's corporate partner, The Ardian Foundation (2022).
Youth Hackathon on Climate Change
Certificate of Achievement for participating in the 2022 Youth Hackathon on Climate Change.
Contact
Get in Touch
I'm always open to research collaborations, data science opportunities, and conversations about forecasting and machine learning. Feel free to reach out — I'd love to hear from you.