NEL ensemble learning for CCN prediction

Knowledge graph centered on NEL ensemble learning for CCN prediction with 60 nodes and 117 connections. Top connected: Temperature, Phytoplankton, Gunnison, Temperature inversion, spectral reflectance.

Method synopsis

Novel ensemble learning approach combining XGBoost, CatBoost, and Random Forest algorithms with SHAP interpretability analysis to predict cloud condensation nuclei from aerosol optical properties.

Synthesized from method descriptions across 1 paper using this protocol.

Procedure as described in the canonical source

Steps below were extracted from the paper that introduces this protocol — Interpretable ensemble learning unveils main aerosol optical properties in predicting cloud condensation nuclei number concentration (2025), NPJ Climate and Atmospheric Science. Implementations in other papers (listed below) may differ.

1. Feature extraction and selection
   Six aerosol optical parameters extracted at three wavelengths (464nm, 529nm, 648nm): absorption coefficient, scattering coefficient, backscattering coefficient, single scattering albedo, backscatter fraction, and Ångström exponent. Recursive Feature Elimination (RFE) combined with manual selection for dimensionality reduction.
   Quantities: Six parameters across three wavelengths as input featuresDuration: Feature engineering applied during preprocessing phaseConditions: Specific wavelengths of 464nm, 529nm, and 648nm selected
   Equipment: RFE algorithms, feature selection algorithms
2. Model development and training
   Three machine learning models trained: XGBoost, CatBoost, and Random Forest. Models combined using ensemble learning approach. Data split into training and testing sets in 8:2 ratio. Bayesian optimization employed for hyperparameter tuning to maximize coefficient of determination (R²).
   Quantities: Three base models combined into ensemble, 80% training, 20% testing splitDuration: Model training performed for each site separatelyConditions: Identical computational conditions for all sites
   Equipment: XGBoost, CatBoost, Random Forest algorithms, Bayesian optimization
3. SHAP interpretability analysis
   SHAP algorithm applied to trained NEL model for feature importance quantification. Individual feature contributions and interaction effects calculated. SHAP values computed across all possible feature combinations to assign importance values and provide interpretable explanations.

Typical Equipment

• XGBoost
• CatBoost
• Random Forest algorithms
• SHAP algorithm
• Bayesian optimization

Output Measurements

• CCN predictions
• feature importance scores
• model performance metrics

Papers Using This Protocol (1)

Species Studied With This Protocol

Places Where Used