Outlier Detection Toolkit
Overview
The Outlier Detection Toolkit applies multiple outlier detection algorithms to your 2D data and compares their results. Visualize your data in a scatter plot with outliers highlighted by each method (Z-score, IQR, Isolation Forest approximation, and Local Outlier Factor), adjust sensitivity thresholds, and see where methods agree or disagree. Perfect for understanding which data points are truly anomalous and how robust your outlier detection is across different approaches.
Tips
- Z-score method works best for normally distributed data - less reliable when distribution is skewed or has heavy tails
- IQR method is robust to extreme values and works well regardless of distribution shape
- Isolation Forest excels at detecting anomalies in multi-dimensional space and can find complex patterns
- Local Outlier Factor identifies points that are outliers relative to their local neighborhood density
- When multiple methods agree on an outlier, it’s likely a true anomaly worth investigating
- Use lower thresholds (more sensitive) for quality control; higher thresholds for exploratory analysis
- Always visualize your data before removing outliers - they might represent valid extreme values or data errors