Electricity: Subsampled Machine Learning Benchmark with 10 Classes and 100 Features

A 2000-row, 100-feature subsample of a larger electricity dataset, created with a random seed of 2. It was generated by Eddie Bergman via a Python script that uniformly sampled rows and columns while preserving up to 10 target classes. The original dataset is licensed as a US public domain work.

Use Cases

• Benchmarking multi-class classification models based on the 10 target classes.
• Feature selection experiments based on the 100 sampled columns.
• Testing model performance on stratified subsamples of larger datasets.
• Comparing algorithm results on a controlled, reproducible subset of energy data.

Strengths

• Controlled subsample size of 2000 rows, facilitating faster experimentation.
• Contains 100 features, providing a moderate-dimensional feature space for analysis.
• Sampling was performed with stratification, likely preserving the original class distribution.
• Derived from a known OpenML dataset (ID 44156), providing traceable provenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is known only for this subsample; the scale of the original dataset is not specified.
• Last update date is unknown; freshness unverified.

Provenance

Source

OpenML dataset 44156 (electricity)

Collection Method

Programmatic subsampling via a custom Python function using NumPy and scikit-learn's train_test_split.