Lupinus Seed Spectra for Sweet/Bitter Classification Using VIS-NIR

VIS-NIR spectral data for whole seeds from seven Lupinus species, classified as sweet or bitter. The dataset was used to evaluate five machine-learning algorithms on reflectance and absorbance data, with performance assessed using F1-score and ROC-AUC. It was authored by Josefa Díaz-Álvarez and last updated on 2026-03-18.

Use Cases

• Train machine learning models for seed classification based on spectral reflectance data.
• Compare the effectiveness of different spectral-transformation techniques for improving discrimination.
• Evaluate resampling methods to address class imbalance in agricultural datasets.
• Develop non-destructive quality control pipelines for legume breeding programs.

Strengths

• Dataset is focused on a specific, non-destructive method for seed classification.
• Performance metrics are provided, with top models achieving F1-scores and ROC-AUC above 92%.
• License is CC-BY-4.0, permitting broad reuse with attribution.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• The dataset is small in file size (92.9 KB), indicating a limited scope.

Provenance

Source

figshare

Collection Method

Data acquired using visible–near infrared (VIS-NIR) spectroscopy on whole seeds.

Freshness

Last updated 2026-03-18 07:38:49; freshness should be verified.