Recidivism & Reentry

Predicting Criminal Recidivism Using Specialized Feature Engineering and XGBoost (NIJ Recidivism Forecasting Challenge, Georgia parolees released 2013) (NCJ 305039, 2021)

46 Data Points 6 Sources 13 Entities Research Date: Sep 6, 2026

This NIJ-funded winning paper applies XGBoost machine learning to predict recidivism among approximately 26,000 individuals released from Georgia prisons on discretionary parole in 2013. The authors achieved best-in-class Brier scores for Year 2 recidivism prediction, with the 'Early Recidivism' feature (Recidivism_Arrest_PrevYear) identified as a key driver of model performance. The paper documents feature importance rankings, model parameters, and performance metrics, providing methodological insights for GPS research on Georgia parolee recidivism forecasting.

Where this research is used

This page is a research collection: individual facts drawn from the sources listed further down, each traced to where it came from. A research topic brings many collections together into one overview of an issue. The facts on this page feed these 4 topics; open one to see how they fit with the rest of the evidence.

Parole & Sentencing

Georgia's parole system has contracted sharply over three decades: the Board of Pardons and Paroles released 5,588 people on parole in FY2025, down 42% from 9,455 in FY2019, and its lifer grant rate has fallen from 21.9% in FY2001–FY2006 to roughly 5% today. Average time served on a life sentence for a serious violent felony has risen from under nine years in 1973 to 29.6 years in FY2025 per the Board's own reporting — 31.1 years by GPS's separate count. The result is a system in which parole eligibility at 7, 14, or 30 years increasingly ends in denial, reconsideration up to eight years later, and release for many only at the maximum sentence date.

12,868 data points

Racial Disparities

Black Georgians are roughly a third of the state's population and well over half of its prisoners, and the gap holds at every measured stage — arrest, charge, plea, sentence, supervision, and exoneration. Georgia's own records show the disparity widest at the deep end: about seven in ten people serving life, and 80 percent of those sentenced to life for offenses committed before age 25. The state's longer record supplies the frame — a late-19th-century convict population that was roughly 90 percent Black against a free population roughly 45 percent Black, and a punishment system that still extracts unpaid labor under a constitutional exception never repealed in Georgia.

2,134 data points

Recidivism & Reentry

Georgia reports one of the lowest recidivism rates in the nation — a 25 to 27 percent three-year felony reconviction rate — but that figure counts only new felony convictions, only within three years, and only among people who survive the window. Georgia's own reconviction rate has climbed every year since 2019, reaching 30.7 percent for the 2022 release cohort even as the state spends more than a billion dollars a year on corrections, releases 12,000 to 16,000 people annually with almost no preparation, and funds vocational education at $172,000 statewide. The evidence on what works is unusually clear — education, vocational training, cognitive-behavioral programming, and transitional beds all correlate with substantially lower recidivism — and Georgia has systematized almost none of it at scale.

12,548 data points

Reform Models & Programs

Georgia operates a statewide cognitive-skills curriculum for parolees, twelve reentry and cognitive programming centers, contracted addiction treatment, step-down units inside its Tier segregation system, and an earned-time credit statute — and the national evidence base for each of those program types is unusually strong. What the record does not contain is current, published, rigorously designed outcome data for most of Georgia's own programs, and the state's direct spending on them amounts to a fraction of one percent of the prison budget. This page sets out what the research verifies, what Georgia actually runs, what it spends, what other states have achieved, and where the evidence stops.

12,091 data points

26,000 Dataset size: ~26,000 Georgia parolees released 2… 0.2 Year 1 model Brier score 0.1837 0.1 Year 2 model Brier score 0.1172 0.1 Year 3 model Brier score 0.0720 0.2 Neural Network 10-fold validation Brier score mea… 0.1 Random Forest 10-fold validation Brier score mean…

Key Findings

The most impactful data from this research collection. Tap a card for the full finding, GPS commentary on why it matters, and its source.

All Data Points

46 verified data points extracted from primary sources.

Dataset size: ~26,000 Georgia parolees released 2013 Statistic

The aggregated dataset provided by the NIJ contains approximately 26,000 individuals released from Georgia prisons on discretionary parole for post-incarceration supervision between January 1st, 2013 and December 31st, 2013.

26,000 individuals

parole demographics facilities

NIJ train/test split proportion 70/30 Methodology note

NIJ split the dataset into a training and test set with a 70/30 proportion.

methodology operations

Data providers: GDCS and Georgia Bureau of Investigation Methodology note

Both the GDCS and the Georgia Bureau of Investigation provided data. The GDCS data included demographics, prison and parole case information, prior community supervision history, and supervision activities. The Georgia Bureau of Investigation provided data on prior criminal history measures such as arrest and conviction episodes prior to prison entry.

methodology operations parole

Recidivism measure definition: new felony or misdemeanor arrest within 3 years Methodology note

GCIC data also provides the recidivism measure, defined as a new felony or misdemeanor arrest episode within three years of parole supervision start date. This recidivism measure includes three dichotomous variables measuring if an individual recidivated in the three-year follow-up period (yes/no) as well as recidivated by time period (year 1, year 2, or year 3).

recidivism methodology parole

Early Recidivism feature added for Year 2 and Year 3 models Methodology note

For Year 2 and Year 3, a feature called Early Recidivism was added. This feature delineated whether the individual already returned to prison. This could only be added to Year 2 training and Year 3 training.

methodology recidivism

Low feature importance variables retained in model Methodology note

There were multiple variables that had relatively low feature importance: Gender, Prior_Conviction_Episodes_PPViolationCharge, Prior_Conviction_Episodes_DomesticViolenceCharges, to name a few. However, because computational speed wasn't a priority for this project, all variables, regardless of statistical significance, were kept in model training.

methodology demographics

Software: Python 3.8 and scikit-learn Methodology note

Preprocessing and model construction were performed on Python 3.8. Preprocessing functions and ML models were imported from the Python library scikit-learn.

methodology operations

Missing values imputed using SimpleImputer Methodology note

Missing values were imputed using the SimpleImputer library.

methodology

XGBoost selected as best performing model Finding

In Round 1 of the competition, we analyzed different model performances using 10 fold cross validation and found that on average XGBoost performed the best.

methodology recidivism

Year 1 model Brier score 0.1837 Statistic

Year 1 model achieved a Brier score of 0.1837 and F1 score of 0.3009.

0.2 Brier score

recidivism methodology

Year 2 model Brier score 0.1172 Statistic

Year 2 model achieved a Brier score of 0.1172 and F1 score of 0.2512.

0.1 Brier score

recidivism methodology

Year 3 model Brier score 0.0720 Statistic

Year 3 model achieved a Brier score of 0.0720 and F1 score of 0.0492.

0.1 Brier score

recidivism methodology

Year 2 model performed very well on hidden dataset Finding

In terms of our performance in the challenge, we found that our Year 2 model performed very well on the hidden dataset when compared with other submissions.

recidivism methodology

Random Forest performed well but more variable than XGBoost Finding

The Random Forest performed well but had a much more variable performance than the XGBoost. Other regression and classification methods were assessed but performances varied widely, none achieving the success of the XGBoost.

methodology recidivism

Class imbalance affected all models Finding

All models suffered from heavy class imbalance due to the relative proportion of the non-recidiviated class to the recidivated class. Synthetic augmentation methods like SMOTE were used to mitigate this class imbalance, but did not provide significant benefits.

methodology recidivism

Optimal probability threshold 0.5 for XGBoost Finding

Different probability thresholds were experimented to optimize XGBoost predictions. With the training and testing data split that was used, optimal brier scores were achieved by using a 0.5 threshold. Increasing or decreasing the threshold leads to worse metrics.

methodology recidivism

Early Recidivism feature significantly improved model performance Finding

More specifically the addition of our Early Recidivism feature (Recidivism_Arrest_PrevYear) played a significant role in helping our model performance improve for the later rounds when compared with other submissions.

recidivism methodology

F1 score recommended for future imbalanced data studies Finding

One metric that could be used in future studies is the F1 score. The F1 score is the harmonic mean of the precision and recall, making it an effective score for delineating the performance of models which were trained on highly imbalanced data.

methodology recidivism

Top 5 features for Year 1 recidivism prediction Finding

Top five most important features for prediction of recidivism in Year 1: Age_at_Release, Prior_Arrest_Episodes_Felony, Gang_Affiliated, Prison_Years, Prior_Arrest_Episodes_Property.

recidivism gangs demographics methodology

Top 5 features for Year 2 recidivism prediction Finding

Top five most important features for prediction of recidivism in Year 2: Recidivism_Arrest_PrevYear, Percent_Days_Employed, Jobs_Per_Year, Age_at_Release, Avg_Days_per_DrugTest.

recidivism employment drugs methodology

Top 5 features for Year 3 recidivism prediction Finding

Top five most important features for prediction of recidivism in Year 3: Recidivism_Arrest_PrevYear, Percent_Days_Employed, Jobs_Per_Year, Age_at_Release, Avg_Days_per_DrugTest.

recidivism employment drugs methodology

Neural Network 10-fold validation Brier score mean 0.1525 Statistic

Neural Network with hyperparameter tuning achieved a mean Brier score of 0.1525 (SD 0.00336) and mean F1 score of 0.7850 (SD 0.0090) in 10-fold validation.

0.2 Brier score

methodology recidivism

Random Forest 10-fold validation Brier score mean 0.1087 Statistic

Random Forest with hyperparameter tuning achieved a mean Brier score of 0.1087 (SD 0.000126) and mean F1 score of 0.8390 (SD 0.00152) in 10-fold validation.

0.1 Brier score

methodology recidivism

XGBoost 10-fold validation Brier score mean 0.0945 Statistic

XGBoost with hyperparameter tuning achieved a mean Brier score of 0.0945 (SD 0.000335) and mean F1 score of 0.8559 (SD 0.00128) in 10-fold validation.

0.1 Brier score

methodology recidivism

Oracle team 1st place Year 2 female parolees Brier score 0.1233 Statistic

In Year 2 female parolees competition, Oracle placed 1st with a Brier score of 0.1233.

0.1 Brier score