rc-coating-ml

Predicting sub-ambient cooling performance of passive surface coatings from formulation parameters using machine learning.

Note: The dataset in this repository is placeholder/demo data that mirrors the structure of the real experimental dataset. The full research data will be released alongside publication. This repo demonstrates the ML pipeline architecture.

Problem

Selecting the optimal filler composition for a passive cooling coating currently requires iterative 24-hour outdoor experiments, and each run tests only a small number of variants. This project builds an ML pipeline to predict sub-ambient surface temperature (DeltaT vs reference) from formulation parameters, reducing the number of experiments needed.

Pipeline

Raw CSV  ->  EDA  ->  Feature Engineering  ->  Random Forest  ->  Experiment Recommendations
  01           01              02                    03

Results (Demo Data)

Metric	Value
CV Strategy	Leave-One-Out (n=19)
R2	0.911
MAE	0.075 degC
RMSE	0.107 degC
Top feature	mean_dT_day

Notebooks

Notebook	Description
01_eda_rc_coating.ipynb	EDA: distributions, 24h profiles, correlations
02_features.ipynb	Feature engineering from time-series + formulation metadata
03_model.ipynb	Random Forest + LOOCV + feature importance + error analysis

Project Structure

rc-coating-ml/
├── data/
│   ├── raw/            # demo temperature CSVs (same structure as production)
│   └── features.csv    # extracted feature matrix
├── notebooks/          # EDA, features, model
├── figures/            # generated plots (gitignored)
└── requirements.txt

Quick Start

python -m venv .venv && .venv\Scripts\activate
pip install -r requirements.txt
jupyter notebook

Run notebooks in order: 01 -> 02 -> 03.

Status

• Environment setup
• EDA notebook (01_eda_rc_coating.ipynb)
• Feature engineering (02_features.ipynb)
• Random Forest + LOOCV (03_model.ipynb)
• Full dataset integration (post-publication)