Polyurethane Glass Transition Temperature Predictions with SHAP Interpretability

A curated dataset for predicting the glass transition temperature (Tg) of polyurethanes using machine learning. The dataset, authored by Yuejie Qin and last updated in April 2026, was used to benchmark algorithms, with an optimized XGBoost model achieving R² > 0.91 and RMSE ≈ 9.6 K. SHAP analysis identified key physicochemical descriptors governing Tg, such as the density of quaternary carbon units in the hard segment.

Use Cases

• Benchmarking machine learning algorithms for materials property prediction based on the described physicochemical descriptors.
• Training predictive models for glass transition temperature based on features like hard segment weight fraction and molecular volume.
• Applying SHAP analysis for mechanistic interpretation of feature contributions to polymer thermal behavior.
• Developing data-driven design rules for polyurethane formulation based on the identified feature groups.

Strengths

• Model performance metrics are explicitly reported (R² > 0.91, RMSE ≈ 9.6 K).
• Dataset is described as carefully curated and expanded for the study.
• Key predictive features are identified and categorized via SHAP analysis.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.

Provenance

Source

figshare, authored by Yuejie Qin.

Collection Method

Likely contains experimental or literature-derived data on polyurethane properties, curated for ML modeling.

Time Range

null

Freshness

Last updated 2026-04-15 06:04:46; freshness should be verified.

Geography

null