QSAR-Biodeg: Molecular Properties for Biodegradability Prediction

A subsampled dataset for quantitative structure-activity relationship (QSAR) modeling of chemical biodegradability. It was created by Eddie Bergman from the original qsar-biodeg dataset on OpenML using a controlled random sampling process. The subsampling parameters included a seed of 4, a maximum of 2000 rows, 100 columns, and 10 classes, with stratification applied.

Use Cases

• Training classification models to predict biodegradability class based on molecular descriptors.
• Benchmarking feature selection algorithms on a controlled subset of chemical properties.
• Developing QSAR models for regulatory assessment of chemical environmental persistence.
• Studying the impact of dataset stratification on model performance for imbalanced chemical data.

Strengths

• Subsampling was performed with a fixed random seed (4), ensuring reproducibility.
• The creation process controlled for maximum rows (2000), columns (100), and classes (10).
• Stratification was applied during row sampling, which likely preserves class distribution.

Limitations

• The exact number of rows, columns, and their specific names are unknown from the metadata.
• Description metadata is limited; actual data quality requires manual inspection after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

OpenML dataset qsar-biodeg (ID 1494), subsampled by Eddie Bergman.

Collection Method

Algorithmic subsampling from a parent dataset using random selection of rows, columns, and classes.