Course contentsShow
← Data Science
Lesson 1210 of 2,14530. Exploratory Data Analysis (EDA) ProcessFree lesson

Relationship Exploration: Correlation and Association

Check correlations between numerics, associations between categoricals, and relationships with target variables.

Relationship Exploration: Correlation and Association

You'll learn how to systematically uncover relationships between variables and with your target variable during EDA.

What This Means

After understanding individual variables and pairs, you need to formally explore how they relate to each other and, crucially, to whatever you're trying to predict or explain. This step identifies which variables move together, which categories associate with outcomes, and which features deserve deeper investigation.

Core Workflow

Numeric-to-Numeric: Use correlation coefficients (Pearson, Spearman) and correlation matrices to spot linear and monotonic relationships. High correlation between predictors might indicate redundancy; correlation with your target suggests predictive power.

Categorical-to-Categorical: Build contingency tables and apply association measures like Cramér's V or chi-square tests. These reveal whether category combinations occur more or less often than chance would predict.

Numeric-to-Categorical: Compare distributions using grouped summary statistics and visualizations (box plots by group). Check if numeric values differ meaningfully across categories.

Everything-to-Target: Always examine relationships with your target variable first. A feature strongly correlated with your target deserves priority, even if it's weakly related to other variables.

Practical Approach

Think of this as detective work: you're looking for "suspects" (variables that matter) and "accomplices" (variables that work together). High correlation doesn't mean causation, but it signals where to investigate. Missing or weak relationships are equally important—they tell you what won't help your analysis.

Document surprising findings: variables you expected to correlate that don't, or unexpected strong associations. These often reveal data quality issues or important domain insights.

Key Takeaway: Systematically checking correlations between numerics, associations between categoricals, and relationships with your target reveals which variables matter most and how they interact—guiding feature selection and hypothesis refinement.