A fully reproducible, production-style machine learning pipeline for predicting molecular melting points using engineered chemical descriptors and tabular feature reduction techniques.
This project focuses on building a scalable and maintainable ML workflow rather than just achieving a single leaderboard score.
It demonstrates how I:
- design end-to-end pipelines
- manage experiments systematically
- think about performance, reproducibility, and collaboration
- Python (core language)
- Pandas, NumPy – data processing
- Scikit-learn – modeling, evaluation, pipelines
- XGBoost, LightGBM – boosting models
- RDKit – chemistry feature generation
- SciPy / Sparse Matrices – efficient memory handling
- Joblib – model persistence
- Matplotlib / Seaborn – visualization
- VSCode + Conda Environment – development environment
✔️ Clean, modular, production-ready project structure
✔️ Automated feature engineering & feature reduction
✔️ K-Fold evaluation with experiment logging
✔️ Model comparison workflow
✔️ Hyperparameter tuning pipeline
✔️ Reproducible scripts (build → train → inference)
✔️ Results tracking + metrics summary CSV
✔️ Designed with “team collaboration” mindset
You can:
- preprocess the dataset
- build molecular descriptors
- reduce high-dimensional features
- train baseline + tuned models
- compare model performance
- reproduce every stage consistently
- extend the pipeline easily