AI Dictionary › AI Fundamentals

Outlier

Outlier (Valore anomalo)

An outlier is an observation that deviates markedly from the typical behavior of the rest of the dataset: a transaction a thousand times larger than average, an age recorded as 150 years, a sensor that for an instant reports a physically impossible value. It can be a measurement or data entry error, or a rare but real and potentially very informative event.

Definition

How it works

There are several methods for detecting outliers, and the choice depends on how well the data's distribution is understood. Classical statistical methods use thresholds based on standard deviation or interquartile range, flagging anything too far from the center of the distribution as anomalous. Density- or distance-based methods, such as isolation forest or local outlier factor, identify points that lie in sparsely populated regions of the feature space relative to their neighbors. The decision of what to do once an outlier is found, whether to remove it, correct it, or study it separately, always depends on context: in a dataset of real estate prices, an extreme value may be an error; in a fraud detection system, it is often exactly what is being sought.

Applications

In applied AI, handling outliers matters for two opposite reasons: on one hand, many algorithms, like linear regression or k-means, are sensitive to extreme values and can be significantly distorted by a few untreated outliers; on the other hand, in fields like fraud detection, cybersecurity and industrial quality control, the system's primary goal is precisely to identify outliers, which represent the rare event to be caught.

History & etymology

The term outlier, literally something that lies outside, entered the statistical vocabulary in the nineteenth century, when mathematicians and astronomers began systematically addressing how to treat experimental observations that clearly deviated from other measurements. Twentieth-century statistics then developed increasingly refined formal criteria for distinguishing a genuine measurement error from a rare but real observation, a distinction that remains at the heart of machine learning practice today.

Related terms

• Missing values
• Standardization
• Data cleaning
• Clustering
• K-Nearest Neighbors (KNN)

More in AI Fundamentals

• LLM (Large Language Model)
• Token
• Context Window
• Temperature
• AI Hallucination
• Fine-tuning
• RAG (Retrieval Augmented Generation)
• Embedding

Put it into practice

• Real scenario: apply this concept with an instant AI verdict.
• Daily gym: short drills to train dictionary concepts.
• Playoff game: challenge other players, elimination-style.

📱 Download the Android app (beta) iOS coming soon

Say what you mean. Get what you need.

Grace Certified, the AI coach that trains and certifies your prompt engineering, by Agora Intelligence.