An exception is an object or signal generated when an error occurs during program execution, which can be caught and managed to prevent abnormal program termination.

Shaimaa Farouk Hello, thanks for the answer. Most of people do indeed regard exceptions as errors.

Yet, since the arrival of "big data" it's becoming almost impossible to distinguish upfront between Error and a true case of an extreme value.

Also, errors may have a meaning... For example, in one of my projects of Cervix Cancer the Physicians used to skip health check questions. I included those records in the analysis and found that "missing value" were a typical characteristic of the "light case" pattern.

Defining an Exception in research and data science requires a balance between methodological rigor and openness to hidden phenomena. Traditional practice often reduces exceptions to “outliers,” which are simply filtered out on the assumption that they are errors or irrelevant. While this approach simplifies analysis, it risks eliminating signals that may hold critical scientific meaning.

With the advent of big data, distinguishing between an error and a true extreme case has become increasingly complex. Anomalous patterns may initially appear as noise, but deeper inspection can reveal that they are either (1) data recording mistakes, (2) rare manifestations of already known mechanisms, or (3) genuinely new patterns that challenge existing models. In some domains, what first looks like an error later proves to be a signature of a meaningful subgroup. For example, missing values in medical records may not be random: they can indicate behavioral or clinical differences between patient cohorts, as seen in the cervical cancer case where skipped health check questions correlated with “light case” outcomes.

Therefore, a refined definition of Exception could be:

An Exception is a pattern of data points or behaviors that cannot be explained by the dominant factors linking the majority of the dataset. Exceptions differ primarily by structural or behavioral mechanisms rather than by numerical magnitude alone.

This perspective avoids the pitfall of carving results to fit preconceived hypotheses. Instead of automatically discarding exceptions as outliers, they should be categorized into:

For me the following definition works best:

Exceptions are patterns with different characteristics, or an event that is indifferent to its pattern's cause-effect factors.