Skip to content
View brianurban's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report brianurban

Block user

Prevent this user from interacting with your repositories and sending you notifications. Learn more about blocking users.

You must be logged in to block users.

Content in all repositories owned by your account will be closed.
Maximum 250 characters. Please don’t include any personal information such as legal names or email addresses. Markdown is supported. This note will only be visible to you.
Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
brianurban/README.md

Hi, I'm Brian! 👋

Physics-trained data scientist teaching machines to handle messy, imbalanced, real-world scientific data—where accuracy tables lie and robustness matters.

MSc Data Science (DePaul University, June 2026) · BSc Physics, Minor in Applied Mathematics (NEIU, May 2021)


🔍 Currently: Actively interviewing for data science / ML engineering roles · Exploring PhD programs in astrophysics (data science × nuclear astrophysics)

🔬 Research Interests: Deep learning applications in astrophysics — NEO hazard classification, nuclear fusion prediction in stars, and model robustness under measurement uncertainty.


🚀 Featured Projects

Jan – Mar 2026

Asteroid hazard classifier built with SMOTE and gap-penalized loss on a heavily imbalanced dataset (12.7% positive rate). Engineered orbital features validated via K-Means clustering, achieving 2.3× enrichment of hazardous classes.

Sept – Nov 2025

ARIMA and Fourier harmonic regression to detect transits in Kepler light curves, with robust MAD dip detection outperforming classical thresholds on confirmed hosts vs. controls in a case-control design.

Sept – Nov 2025

EfficientNet-B0 CNN fine-tuned for 10-class galaxy classification on 17,736 DECaLS images via transfer learning and weighted loss — 83% test accuracy vs. 15% for classical baselines.

Apr – Jun 2025

MLP, logistic regression, and decision trees compared across six Gaussian noise levels using 5-fold CV. Key finding: peak clean-data accuracy is a misleading model selection metric for noisy scientific data.


🛠️ Tech Stack

Core stack: Python · R · SQL · PyTorch · scikit-learn · Pandas / NumPy · Docker · AWS / Azure

Also working with: JavaScript / React · OpenCV · Albumentations · SciPy · ggplot2 · forecast · Jupyter · Tableau · LaTeX · Git


💼 Experience

📚 Math and Physics Tutor · Malcolm X College · Nov 2024 – Present

Developed data-informed learning frameworks, delivering workshops on quantitative and statistical methods to students of varying technical fluency.

⚙️ Electro-Mechanical Survey Technician · Argonne National Laboratory · Feb – Sept 2023

Managed QA data for 700+ magnet assemblies on the Electron-Ion Collider upgrade project, applying physics-informed diagnostics and building structured data pipelines for cross-team traceability.

💻 Full-Stack Software Developer · Revature · May – Sept 2022

Built predictive analytics modules and interactive data dashboards, deploying scalable Azure cloud solutions and React-based visualization interfaces for real-time decision-making.


⭐️ Beyond the Code

  • 🔭 I bring my telescope to local star parties and share the night sky through astrophotography
  • 🎹 Piano, choir singing, art, and cooking
  • 🎮 When I'm not stargazing: Metal Gear Solid, Kingdom Hearts, and Final Fantasy
  • 🚴‍♂️ Cycling Chicago's Lakefront Trail

Popular repositories Loading

  1. stellar-classification-comparative-analysis stellar-classification-comparative-analysis Public

    Comparative analysis of ML model robustness to measurement noise on stellar classification using PyTorch, scikit-learn, and 5-fold cross-validation.

    Jupyter Notebook

  2. brianurban brianurban Public

  3. neo-hazard-classification neo-hazard-classification Public

    Comparative analysis of 12 machine learning models (deep learning + classical ML + ensembles) for near-Earth object (NEO) hazard classification on imbalanced orbital data (338K records, 6.84:1 ratio).

    Jupyter Notebook

  4. galaxy-morphology-classification galaxy-morphology-classification Public

    Comparative study of EfficientNet-B0 transfer learning vs. traditional ML (SVM, RandomForest, ExtraTrees, kNN) on 28 hand-crafted morphological features for galaxy classification using the Galaxy10…

    Jupyter Notebook

  5. exoplanet-transit-detection exoplanet-transit-detection Public

    A four-star case-control study in R to detect exoplanet transits in Kepler light curves via ARIMA and Fourier harmonic regression with robust MAD dip detection.

    Jupyter Notebook