Pair Plot (Scatter Plot Matrix)

Purpose

Visualize pairwise relationships between all numerical variables in a dataset simultaneously. Essential for initial exploratory data analysis. Used for Collinearity and Interaction detection.

Analysis Type

Multivariate

What to Look For

1. Correlation Patterns

• Strong linear patterns: Highly correlated variables (potential multicollinearity)
• Random scatter: Low or no correlation
• Curved patterns: Non-linear relationships

2. Diagonal Plots

• Show univariate distributions (histograms or KDE)
• Check for normality, skewness, and outliers

3. Redundant Features

• Variables with near-perfect linear relationships
• Candidates for removal to reduce dimensionality

4. Feature Selection

• Variables showing strong relationships with target variable
• Variables with no patterns may be less useful

5. Data Quality

• Outliers visible across multiple plots
• Unusual patterns indicating data issues

6. Class Separation

• When using hue parameter with target variable
• Look for variables where classes separate well

7. Linearity

• Linear:
  • Most of the mini scatter plots show elongated, straight ellipses rather than curves.
  • If you see a consistent straight trend across multiple feature pairs, linear structure is likely.
• Non-Linear:
  • You spot curved bands (U, S), "funnel" or "fan" shapes or distinct circular or banana-shaped "clouds" with no direction.
  • Clusters appear (separate blobs), suggesting the relationship depends on subgroup or a non-linear boundary.

Code Example

# Basic pair plot
sns.pairplot(df)
plt.show()

# Pair plot with color by target variable
sns.pairplot(df, hue='target_variable', diag_kind='kde')
plt.show()

# Pair plot for specific columns
sns.pairplot(df[['var1', 'var2', 'var3', 'target']], hue='target')
plt.show()