Outlier Detection

Identify statistical outliers in numeric fields using z-score or MAD methods.

Overview

Outlier detection identifies features with values that are statistically unusual compared to the dataset distribution.

Methods

Z-Score Method

Uses mean and standard deviation:

Z-score = (value - mean) / standard_deviation
Features with |z-score| > threshold are outliers
Sensitive to outliers in calculation

MAD Method

Uses median and median absolute deviation:

Modified z-score = 0.6745 * (value - median) / MAD
Features with |modified z-score| > threshold are outliers
More robust to outliers in calculation

Inputs

Dataset: Any dataset with numeric field
Value Field: Numeric field to analyze
Method: “zscore” or “mad” (default: “zscore”)
Threshold: Z-score threshold or MAD multiplier (default: 2.0)

Outputs

New dataset containing:

Original features
Outlier Score: Z-score or MAD score
Is Outlier: Boolean flag
Original attributes

Example

{
  "dataset_id": 123,
  "value_field": "income",
  "method": "zscore",
  "threshold": 2.0
}

Background Jobs

This analysis runs as a background job. See Outlier Analysis Worker for details.

Use Cases

Data quality assessment
Anomaly detection
Error identification
Extreme value analysis

Notes

Null values are excluded from calculations
Threshold of 2.0 identifies ~5% of data as outliers (normal distribution)
MAD method recommended for skewed distributions
Consider spatial context when interpreting results