Testing Strategies to Improve Fairness for Predictions Made by Machine Learning-Based Gambling Harm Detection Systems
Bibliographic record
Abstract
Researchers in the field of gambling studies have worked to develop machine learning-based harm detection systems that ingest online gambling behaviour data and generate scores that predict the likelihood that a given user will experience some harm-relevant outcome (self-reporting gambling problems, enrolling in a voluntary self-exclusion program, etc.). However, all studies in this area have used a limited set of metrics to assess the ‘overall’ performance of these machine learning models. Given that membership in different sociodemographic groups correlates with individuals’ risk for experiencing various gambling-related harms (Potenza et al., 2019), it is worth investigating whether current practices are developing models that generate equitable predictions for members of different sociodemographic groups, and whether these practices could be adjusted to improve outcome equity. In this study, we will examine an existing dataset of approximately 20,000 users of a provincially-operated gambling website in Canada. Within these data, we will develop three machine learning models that ingest data related to individual users’ transactions (deposits, withdrawals, and use of promotional offers), and generate predictions about whether each user is likely to have reported experiencing a high rate of gambling-related problems over the 12 months in which all data were collected. The results of these three models will be compared both in terms of their overall performance, and their performance within groups with different ages and sexes. We hypothesize that self-reported status as a high-risk gambler (PGSI 8+) will differ by age and sex, consistent with previous research. Further, we hypothesize that overall and group-level classification performance will differ between models built to satisfy fairness via unawareness, classification parity, and outcome calibration.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.004 | 0.008 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.001 | 0.007 |
| Science and technology studies | 0.002 | 0.001 |
| Scholarly communication | 0.005 | 0.001 |
| Open science | 0.003 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".