Mineral Identity Governs Thermal Stability: Gradient-Boosting Prediction of Degradation Parameters in Mineral-Filled Epoxy and Vinyl Ester Composites
Description
This dataset contains 317 curated literature records of thermogravimetric analysis (TGA) data for mineral-filled epoxy and vinyl ester composites, assembled to train and validate gradient-boosting machine learning models (XGBoost, CatBoost) for predicting thermal degradation parameters. DATASET CONTENT: - Compositional descriptors: polymer matrix type (epoxy/vinyl ester), mineral filler identity and formula, filler loading (wt.%), chemical modifier name/type/loading/SMILES - TGA experimental conditions: atmosphere (N2/Air/Argon), heating rate (where reported) - Target variables: onset degradation temperature T5% (°C), peak degradation temperature Tmax (°C), residual mass (wt.%), residue evaluation temperature - Engineered features: Magpie elemental descriptors (145 features per mineral), RDKit molecular fingerprints for modifiers METHODOLOGY: Data extracted from peer-reviewed publications (2015-2024) with standardized normalization of experimental conditions. Group-aware train/holdout splitting (80/20) based on matrix-mineral-modifier combinations to prevent data leakage. Used in: "Mineral Identity Governs Thermal Stability: Gradient-Boosting Prediction of Degradation Parameters in Mineral-Filled Epoxy and Vinyl Ester Composites" (Polymer Testing). FILES: - TGA_dataset_317_records.csv (raw curated data) - Feature_engineered_dataset.csv (with Magpie and RDKit descriptors) - Data_dictionary.txt (column descriptions) CITATION: If using this dataset, please cite the original article: [DOI will be added upon publication]