Machine Learning Dataset for Copper Plating Leveler Discovery

A 2026 study by Bo Yang presents a dataset for predicting the performance of acidic copper plating levelers. It includes experimental Dissolution Peak Decrease Amount (DPDA) values and DFT-calculated adsorption energies (E_ads) as targets, with 24 theoretical molecular properties as features. The framework was used to screen 29,785,186 compounds and identify five novel levelers.

Use Cases

• Train an XGBoost Regression model on the 24 theoretical molecular properties to predict Dissolution Peak Decrease Amount (DPDA) values.
• Perform SHAP analysis on the 24 molecular descriptors to identify key features governing leveling efficacy and structure-activity relationships.
• Use the DFT-calculated adsorption energies (E_ads) as a target variable for screening candidate molecules from large compound libraries.
• Build a Classification and Regression Tree model to classify leveler performance based on the provided molecular property features.

Strengths

• Dataset is derived from a published scientific study proposing an integrated ML and DFT framework.
• Features include 24 distinct theoretical calculation molecular properties for model training.
• The predictive models were applied to screen a massive database of 29,785,186 compounds.

Limitations

• The exact number of rows in the core training dataset is unknown, limiting assessment of model training scale.
• Data is static and tied to a specific research publication, with no indication of ongoing updates or expansion.
• The dataset's utility is highly specialized for computational chemistry and electroplating, requiring domain expertise.

Provenance

Source

figshare, authored by Bo Yang.

Collection Method

Experimental DPDA values from literature combined with Density Functional Theory (DFT) calculations.

Time Range

null

Freshness

Last updated March 2026.

Geography

null