CPU Activity Regression Benchmark

## Overview The cpu_act dataset is a classic regression benchmark containing 8,192 observations of computer systems activity measured on a multi-user Sun Sparcstation 20/712 with 128 MB RAM. Each row captures a snapshot of system performance counters, and the target is the portion of time CPUs run in user mode (`usr`). The dataset has 22 numeric columns (21 features + 1 target) and is distributed as ARFF via OpenML. ## Schema - `lread` — numeric — reads (transfers per second) between system memory and user memory - `lwrite` — numeric — writes (transfers per second) between system memory and user memory - `scall` — numeric — number of system calls per second - `sread` — numeric — number of system read calls per second - `swrite` — numeric — number of system write calls per second - `fork` — numeric — number of system fork calls per second - `exec` — numeric — number of system exec calls per second - `rchar` — numeric — characters transferred per second by system reads - `wchar` — numeric — characters transferred per second by system writes - `runqsz` — numeric — process run queue size - `freemem` — numeric — number of memory pages available to user processes - `freeswap` — numeric — number of disk blocks available for page swapping - `usr` — numeric — TARGET: portion of time (%) CPUs run in user mode - +9 more columns (pgout, ppgout, pgfree, pgscan, atch, pgin, ppgin, pflt, vflt) ## Sources - OpenML dataset 197: https://www.openml.org/d/197 — Public Domain (CC0) - Originally from the Delve repository (University of Toronto) ## Methodology Data was collected from a Sun Sparcstation 20/712 with 128 MB RAM running in a multi-user university environment. System activity was sampled using standard Unix performance counters at regular intervals, producing 8,192 snapshots of paging, system call, I/O, and memory utilization metrics alongside the resulting user-mode CPU utilization. No additional normalization is applied by OpenML beyond ARFF packaging. ## Known gaps & limitations The data reflects a single Sun Sparcstation workstation circa the early-to-mid 1990s and does not generalize to modern multi-core or virtualized hardware. The workload mix is specific to the host university's user population at collection time. Source does not document inter-sample independence or sampling cadence in detail; buyers should validate empirically before drawing causal conclusions. ## Intended use & out-of-scope - IS for: regression benchmarking, tabular ML evaluation, feature-selection studies, teaching examples, and reproducibility comparisons against published baselines. - NOT for: modeling contemporary cloud or multi-core CPU behavior, capacity planning on modern systems, or any production scheduling decisions. Original supplier listing: CPU Activity (cpu_act) Classic regression benchmark predicting CPU user-mode utilization from 21 system activity measures collected on a Sun Sparcstation. 8,192 rows, widely used in ML evaluation.