tax-fraud-xai
Boosted-tree and neural model with SHAP explanations that flags fraudulent filings, on synthetic data.
What it does
Research code that builds a synthetic US tax-return dataset, trains an XGBoost model, an attention neural network and a hybrid of both to score returns for fraud risk, and explains the scores with SHAP plots and attention heatmaps, with a score and action per return. Runs as a series of Python scripts (3.8+). Synthetic data only, marked NOT REAL, so its detection rates are on that data; MIT.
Where it fits on the map
Not on the map: inactive and a demo.
The map places maintained projects for a company tax department or a tax authority, without a demo or paid-service flag. Everything else stays on the list with its label.
Similar projects
Of the same kind, for a shared jurisdiction or tax area. Not a ranking.
- amazon-state-sales-and-tax-summaryTurns marketplace order exports into state-by-state sales and tax totals for US sales-tax returns.
- caseflowTariff classification with legal reasoning for Nigeria's ECOWAS common external tariff and the US HTS.
- CFO StackBeancount accounting run through Claude Code, with jurisdiction packs for Canada, the US, Taiwan and China.For individual taxpayers
- DaLI for RP2Data loader that turns exchange and wallet records into input files for the RP2 crypto tax calculator.For individual taxpayers
- fatca-ides-phpPHP library that converts bank client data into FATCA XML packages for the IRS IDES gateway.
- hs-code-lookup (AccioWork)HTML tool to find 6-digit HS and 10-digit HTS codes for 15 or more product categories, with 2026 rates.
Maintain this project?
Link to this page from your README, so readers find the facts, the notes and where it fits. The badge is a small SVG served from this site.
[](https://tax-oss.com/projects/tax-fraud-xai/)