Outlier

An outlier is a value strikingly far from the rest of the data — far enough that it may be a measurement error or a genuinely exceptional behavior.

An outlier is a fork in the road, not garbage: before removing it you must ask whether the number is wrong or real. A multi-hour session that is really a forgotten browser tab is an error and should be set aside; but the same session, if it is a real user who read content for hours, is a valuable signal, not an outlier. The main danger is that an outlier swallows the mean and standard deviation; one extreme value can skew the whole picture. There are two wrong reactions: thoughtlessly deleting any odd number (which may throw away your most important users), and ignoring them entirely (which makes the mean meaningless). The right path is to use robust measures like the median and percentiles, which inherently resist outliers, and to inspect the outliers themselves separately.