Outlier
In one sentence An outlier is a data point that sits far from the pattern the rest of the data follows — sometimes an error, sometimes the discovery.
Updated
An outlier is a data point that lies unusually far from the rest — it does not fit the pattern the other points form.
In a class photo of forty schoolchildren, the 6-foot-4 teacher standing among them is the outlier. Any "average height in this photo" you compute is now misleading, and any clothing size you order based on that average fits nobody. One extreme point, honest or not, can drag summary numbers badly.
The first question about an outlier is always why it exists, because the answer decides everything:
a typo ₹4,50,000 salary entered as ₹45,000,000 → fix or drop
a measurement fault sensor stuck at 999 → drop
a real rare event one customer who genuinely spends 100× → keep, and think
the actual target the fraud, the failing machine → the point of the projectDeleting outliers by reflex is a classic mistake — in fraud detection and fault monitoring, the outliers are the signal, and a whole field (anomaly detection) exists to find them.
For modelling, sensitivity varies. Squared-error losses and linear-regression chase outliers hard; tree-based models shrug at extreme input values; k-means gets its centres dragged. Common detection tools: distance beyond 1.5 × IQR (the boxplot rule), z-scores, isolation forests, and points that fit no cluster.
Where to go next
- Full lesson: Unsupervised learning
- Related terms: clustering, data-drift, feature-scaling, loss-function