V
The Ledger Beneath the Tide
TaxLens
Which tax returns deserve a closer look? Not the largest, and not the rarest kind — the ones that don’t look like their peers. TaxLens is a Master of Data Science capstone: an unsupervised anomaly-detection model trained on the ATO 2022–23 individual sample file. It sorts taxpayers into peer groups, scores each return against its own group, and explains every score in the return’s own items.
321,890
tax returns, ATO 2022–23 sample
8
peer groups found by K-Means
6,442
flagged — the 2% least like their peers
3
named items explain every flag
A flag is a statistical signal for human review — never evidence of anything.