GreenML: Energy consumption measurements for 8 ML classifiers on CPU-only consumer hardware (600 runs, 3 datasets)

Energy consumption measurements for 8 ML classifiers (Random Forest, XGBoost, SVM-RBF, KNN, Logistic Regression, Decision Tree, MLP, and Naive Bayes) across three datasets (synthetic, Adult Income, and Credit Card Fraud) and five dataset sizes (2,000–80,000 rows), with 5 repeats per configuration, yielding 600 total training runs. Energy was measured using AMD μProf (v5.2.431.0) on native Windows 11 (AMD Ryzen 9 8945HS), reading the socket0-package-power counter at 100 ms intervals. This is hardware-level CPU package power measurement, not a TDP-based proxy. Columns: run_id: unique identifier per run duration_sec: single-fit wall-clock time (seconds) kwh: energy per fit (kWh, measured) avg_watts: mean CPU package power during training (W) power_samples: number of 100ms power samples captured n_repeats: loop repetitions used for sub-second models Dataset: synthetic / adult / creditcard model: model name size: training dataset size (rows) n_features: number of input features repeat: repeat index (0–4) accuracy: training-set accuracy Associated paper: GreenML: Pre-Training Energy Consumption Prediction for Machine Learning Models on CPU-Only Consumer Hardware (submitted 2026).