metadata

language:
  - en
tags:
  - compas
  - tabular_classification
  - binary_classification
  - UCI
pretty_name: Compas
size_categories:
  - 1K<n<10K
task_categories:
  - tabular-classification
configs:
  - encoding
  - two-years-recidividity
  - two-years-recidividity-no-race
  - priors-prediction
  - priors-prediction-no-race
  - race
license: cc

Compas

The Compas dataset for recidivism prediction. Dataset known to have racial bias issues, check this Propublica article on the topic.

Configurations and tasks

Configuration	Task	Description
encoding		Encoding dictionary showing original values of encoded features.
two-years-recidividity	Binary classification	Will the defendant be a violent recidivist?
two-years-recidividity-no-race	Binary classification	As above, but the `race` feature is removed.
priors-prediction	Regression	How many prior crimes has the defendant committed?
priors-prediction-no-race	Binary classification	As above, but the `race` feature is removed.
race	Multiclass classification	What is the `race` of the defendant?

Usage

from datasets import load_dataset

dataset = load_dataset("mstz/compas", "two-years-recidividity")["train"]

Features

Feature	Type	Description
`sex`	`int64`
`age`	`int64`
`race`	`int64`
`number_of_juvenile_fellonies`	`int64`
`decile_score`	`int64`	Criminality score
`number_of_juvenile_misdemeanors`	`int64`
`number_of_other_juvenile_offenses`	`int64`
`number_of_prior_offenses`	`int64`
`days_before_screening_arrest`	`int64`
`is_recidivous`	`int64`
`days_in_custody`	`int64`	Days spent in custody
`is_violent_recidivous`	`int64`
`violence_decile_score`	`int64`	Criminality score for violent crimes
`two_years_recidivous`	`int64`