Machine Learning Optimizer for Carbon Nanotube Dispersions Across 666 Formulations

A dataset of 666 carbon nanotube dispersions systematically varying 36 organic dispersants, 22 solvents, and two dispersion methods, created by Hirokuni Jintoku and last updated in April 2026. It includes quantified dispersion quality (D_score) and structural integrity (I_G/I_D) for each formulation. The data was used to train an XGBoost model for screening-grade formulation design.

Use Cases

• Training ML models to predict CNT dispersibility based on molecular descriptors and experimental variables.
• Screening solvent-dispersant combinations for high-quality, stable CNT dispersions.
• Analyzing the impact of process intensity on CNT structural integrity (I_G/I_D).
• Virtual screening within the learned domain to identify improved formulations.
• Extending the prescreening framework to other nanomaterial dispersion systems.

Strengths

• Systematically covers a combinatorial design space with 666 distinct dispersions.
• Includes two key target metrics: D_score for dispersion quality and I_G/I_D for structural integrity.
• Model performance metrics are provided (R²=0.57 for D_score, R²=0.73 for I_G/I_D).
• Feature importance analysis (SHAP) reveals governing factors for each target.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• The model showed limited quantitative accuracy for dispersants outside the training set.

Provenance

Source

figshare

Collection Method

Experimental investigation of CNT dispersions with systematic variations in composition and processing parameters.

Time Range

null

Freshness

Last updated 2026-04-09 17:37:28; freshness should be verified.

Geography

null