Exploratory data analysis

Exploratory data analysis (EDA) is the open-ended first pass an analyst makes through a new dataset before running any formal model or test: looking at distributions, checking for missing or duplicated values, spotting outliers, and getting a feel for relationships between variables. Its goal is understanding and hypothesis generation, not a final answer.

In practice, EDA relies heavily on summary statistics and data visualization—histograms, scatter plots, box plots—because visual inspection reveals structure such as skew, clusters, or unexpected gaps that a single summary number can hide. It overlaps with descriptive analytics but is more exploratory and less polished: an EDA chart is meant for the analyst's own understanding, not a finished report.

EDA matters because it catches problems early: a data quality issue found during exploration is far cheaper to fix than one discovered after a model has already been built on bad data, and a distribution that turns out to be highly skewed or bimodal often changes which statistical method or transformation is appropriate before running regression analysis or any other formal test. The common pitfall is treating exploratory patterns as confirmed findings—EDA is meant to generate hypotheses to test formally, and mining a dataset for patterns without a pre-specified hypothesis raises the risk of finding spurious relationships that do not hold up on new data.