DataTerminology.comNEW
Also known as: Anomalies
An observation that deviates markedly from other observations, potentially indicating variability, errors, or rare events.
DESCRIPTION
Outliers are data points that differ substantially from the bulk of observations. They may signal data-entry errors, distributional tails, novel phenomena, or system issues. Analysts detect outliers using robust statistics (e.g., IQR rules, median absolute deviation), distance-based methods, or model-based approaches. Decisions about removing or retaining outliers depend on domain context and analysis goals. Proper handling improves model robustness, fairness, and interpretability.
ALTERNATIVE DEFINITIONS
NIST/SEMATECH e-Handbook of Statistical Methods
An outlier is an observation that appears to be inconsistent with the remainder of that set of data.
Stresses inconsistency relative to the dataset as a whole.
scikit-learn
Outlier detection refers to the problem of finding patterns in data that do not conform to expected behavior.
Frames outliers as deviations from expected patterns, motivating algorithmic detection.
RELATED TERMS
Data Anomaly Detection Robust Statistics Z-score Interquartile Range Mahalanobis Distance Scrubbing Winsorization Box Plot Data Profiling Data Quality
COMPETING TERMS
Inlier
KNOWLEDGE AREAS
Data Science Mathematics Data Quality Testing & Validation
TARGET AUDIENCE
Data Scientist Data Analyst Data Engineer Data Steward MLOps Engineer
EXTERNAL SOURCES
- Wikipedia - Outlier
- scikit-learn - Overview of outlier detection methods
- NIST - NIST guidance on outliers
TOP QUESTIONS
Why are Outliers important in data analysis?
Outliers can distort statistical estimates and model performance but may also reveal rare events or errors. Recognizing them is crucial for robust, trustworthy insights.
Category: Relevancy
How do I detect Outliers?
Use robust measures like IQR or MAD, z-scores for near-normal data, or algorithmic detectors (Isolation Forest, LOF) for complex patterns.
Category: Requirements
Should I remove Outliers?
It depends on context. Remove clear errors; otherwise, use robust models or transformations. Document all decisions for reproducibility.
Category: Opinion
What is a common mistake with Outliers?
Automatically dropping extreme values without understanding causes or impacts can bias results and hide important phenomena.
Category: Mistakes
How do Outliers affect machine learning models?
They can skew parameter estimates, degrade generalization, and mislead evaluation. Regularization, robust losses, and preprocessing help mitigate effects.
Category: Risks
What is a robust alternative to mean and standard deviation for Outliers?
Use the median and median absolute deviation (MAD), which are less sensitive to extremes.
Category: Alternatives
How do Outliers relate to data quality?
Some outliers stem from data entry or process errors. Profiling and governance practices help identify and rectify such issues.
Category: Relevancy
Can Outliers indicate model drift in production?
Yes. A sudden increase in outliers may signal upstream changes, sensor faults, or concept drift requiring investigation.
Category: Example
What visualization helps spot Outliers?
Boxplots, scatterplots, and residual plots quickly reveal extreme values and patterns.
Category: Usage