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

# Anomaly detection

> Find outliers, fraud, and quality issues with statistical and ML methods.

Sourcetable offers multiple anomaly detection methods. The AI selects the best approach based on your data and goal.

## Statistical methods

### Z-score

```
"Flag all transactions with a Z-score above 3"
```

Identifies values more than N standard deviations from the mean. Works best with normally distributed data.

### IQR method

```
"Find outliers in the salary column using the IQR method"
```

Flags values below Q1 - 1.5×IQR or above Q3 + 1.5×IQR. More robust to non-normal distributions than Z-score.

### Modified Z-score (MAD)

Uses Median Absolute Deviation instead of standard deviation. Resistant to outliers in the reference data.

## ML methods

### Isolation Forest

```
"Run Isolation Forest to detect anomalous transactions"
```

Isolates anomalies by randomly partitioning the feature space. Works well with:

* High-dimensional data
* No need for labeled examples
* Mixed feature types

### DBSCAN

```
"Find anomalous data points using DBSCAN clustering"
```

Points that don't belong to any cluster are flagged as anomalies. Good for:

* Spatial data
* Clusters of arbitrary shape
* Unknown number of anomalies

### Local Outlier Factor (LOF)

```
"Detect local outliers in the sensor data using LOF"
```

Compares local density of each point to its neighbors. Catches anomalies that are normal globally but unusual locally.

## Time series anomalies

```
"Flag days where website traffic is anomalously high or low"
```

Methods:

* **Rolling statistics** — flag values outside rolling mean ± N×rolling std
* **Seasonal decomposition** — flag large residuals after removing trend and seasonality
* **Prophet-style** — detect changepoints and outliers in seasonal time series

## Use cases

| Domain        | What to detect          | Suggested method           |
| ------------- | ----------------------- | -------------------------- |
| Finance       | Fraudulent transactions | Isolation Forest, Z-score  |
| Manufacturing | Defective products      | LOF, IQR                   |
| IT operations | Server anomalies        | Rolling statistics, DBSCAN |
| Healthcare    | Abnormal lab results    | Modified Z-score           |
| E-commerce    | Unusual order patterns  | Isolation Forest           |
| IoT           | Sensor malfunctions     | Time series methods        |