Definition
EDA uses descriptive stats, histograms, correlation views, and spot checks to learn what the data contains before you commit to a model.
It surfaces leakage risks, imbalance, and weird outliers early.
In simple terms
Walking a property before remodeling — measure rooms, note water damage, then plan the renovation.
Where you see it
- Pandas profiling before a Kaggle baseline.
- Checking token length distributions for an NLP corpus.
How it works
1.Summarize
Shapes, types, missingness, basic stats.
2.Visualize
Distributions and relationships.
3.Hypothesize
Notes that guide features and models.
Why it matters
- EDA prevents weeks of modeling on misunderstood data.
Often confused
EDA is optional once you have deep learning.
Deep models still fail on silent data bugs EDA would catch.