Data format
What to send SymNexusSentinel
The inputs the platform expects, and the labels it learns from.
Input
Input format
Rows of activity with any mix of numeric and categorical fields: transactions, logins, sensor windows, invoices.
| transaction_id | amount | hour | distance_from_home_km | merchant_category | card_present |
|---|---|---|---|---|---|
| 884201 | 23.10 | 8 | 2.4 | transport | 1 |
| 884202 | 1450.00 | 3 | 7200 | online | 0 |
transaction_id,amount,hour,distance_from_home_km,merchant_category,card_present
884201,23.10,8,2.4,transport,1
884202,1450.00,3,7200,online,0For training
Label format
Optional. If you already know some confirmed cases (1 = fraud, 0 = normal), add them to measure detection quality; they're never required to run.
| transaction_id | is_fraud |
|---|---|
| 884201 | 0 |
| 884202 | 1 |
Tips for good results
- Mostly-normal history works best; a small share of anomalies is expected.
- Include the fields an analyst would look at when investigating.
- Set the expected alert rate to control volume.