Kansas Well Logs for Sonic and Gamma Ray Prediction via KNN

A machine learning framework achieved test R² values from 0.942 to 0.963 for predicting missing Sonic logs and 0.927 to 0.930 for Gamma Ray logs. This dataset from the University of Kansas supports a study on using K-Nearest Neighbors regression to address gaps in geophysical well data. The workflow involves correlation-guided feature selection and min–max normalization on data from five wells.

Use Cases

• Predict missing Sonic (DT) log values based on other available well log measurements.
• Predict missing Gamma Ray (GR) log values for enhanced stratigraphic interpretation.
• Benchmark non-parametric regression methods like KNN against other imputation techniques for geophysical data.
• Study the impact of Pearson correlation-based feature selection on model performance in data-limited scenarios.

Strengths

• High predictive performance demonstrated with test R² values up to 0.963.
• Workflow is documented with specific steps: 70/30 train-test split and RMSE optimization for selecting neighbors.
• Dataset is openly available under a CC-BY-4.0 license.

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.
• Row count is unknown, which may limit suitability assessment.

Provenance

Source

Open-source dataset from the University of Kansas.

Collection Method

Data likely gathered from geophysical well logging tools.

Freshness

Last updated 2026-04-21 05:25:21; freshness should be verified.

Geography

Focus on five wells, likely in Kansas, USA.