단변량 분석 후 이변량 분석

Good EDA has an order. First understand each column alone, then study how pairs move together. Start univariate, finish bivariate. 📊

Looking at a single column on its own is univariate analysis. You ask what is typical and how spread out the values are.

For one numeric column, a histogram reveals its shape: bell, skewed, or lumpy. Shape hints at how to model it later.

For a single label column, a bar chart of value_counts shows which categories dominate and which are rare.

Each numeric column has a center, like the mean, and a spread, like the standard deviation. Together they summarize it in two numbers.

Once columns make sense alone, study pairs. Bivariate analysis asks whether two variables rise and fall together.

For two numeric columns, a scatter plot shows whether they trend together, oppose, or scatter with no link at all.

To compare a number across groups, group then summarize. A grouped mean shows how charges differ by plan in one line.

For two label columns, a crosstab counts every combination, so you can see which pairs appear together often.

A scatter hints, but a number confirms. The correlation between two numeric columns measures how tightly they move together.

Jumping to relationships before knowing each column hides traps. The univariate-first order keeps your conclusions trustworthy.

You want to see if two numeric columns move together.

You explore one column with histograms and bars, then pairs with scatters, groupings, and correlation. Order keeps insights honest. ✅