> ## Documentation Index
> Fetch the complete documentation index at: https://docs.kairoconnect.com/docs/llms.txt
> Use this file to discover all available pages before exploring further.

# SHAP features

> Interpreting top_features in scoring responses

Each scoring response includes up to **5** highest-impact SHAP features in `top_features[]`:

```json theme={null}
{
  "top_features": [
    {"feature": "amount_log", "importance": 0.05},
    {"feature": "channel_ord", "importance": -0.03}
  ]
}
```

| Field | Description |
| - | - |
| `feature` | Derived model feature name |
| `importance` | Signed SHAP contribution. Positive increases fraud score; negative decreases it |

## Common features

| Feature | Meaning |
| - | - |
| `amount_log` | log1p(amount) — compresses large amounts |
| `amount_zscore` | Amount deviation vs user history |
| `tx_count_1h` | User transactions in the last hour |
| `tx_count_24h` | User transactions in 24 h |
| `channel_ord` | Ordinal encoding of payment channel |
| `category_ord` | Ordinal encoding of MCC category |
| `transaction_type_ord` | Ordinal encoding of transaction type |
| `hour_of_day` | UTC hour (0–23) |
| `is_night` | Transaction between 00:00–06:00 UTC |
| `amount_x_night` | Amount × night-time interaction |

The model uses dozens of features derived from your request and historical behavior (transaction velocity, user and merchant patterns). Exact names may evolve between model versions; `model_version` in the response shows the product label (`kairo-shield`).

> [!NOTE]
>
> Use `top_features` for operational explainability (why a transaction was flagged), not as input fields for future requests.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.