Multivariate Analysis

Discover hidden patterns by analyzing multiple variables simultaneously to uncover relationships that single-variable analysis completely misses.

The Multiple Variable Reality

Bivariate analysis showed you two-variable relationships. But business decisions rarely depend on just two factors. When Flipkart wants to predict customer lifetime value, they consider age, purchase history, city, product preferences, and seasonal behavior all together. That's where multivariate analysis becomes your power tool.

Think of it like understanding traffic patterns. Single variable: "Rush hour has more cars." Bivariate: "Rain increases traffic time." Multivariate: "Rush hour + rain + Friday + cricket match = complete gridlock." Each additional variable changes the story.

Correlation Matrix: The Relationship Map

Your first multivariate tool is the correlation matrix. This shows how every numeric variable relates to every other numeric variable. Honestly, this is underrated — most analysts skip straight to modeling without understanding their data relationships first.

# Import libraries for multivariate analysis
import pandas as pd
import numpy as np
import seaborn as sns

Libraries imported successfully

The scenario: Myntra's analytics team needs to understand what drives high-value orders across their customer segments. The CMO wants insights within 2 hours for tomorrow's strategy meeting.

# Load and explore the dataset structure
df = pd.read_csv('dataplexa_ecommerce.csv')
print("Dataset shape:", df.shape)
print("\nNumeric columns for correlation:")
numeric_cols = df.select_dtypes(include=[np.number]).columns.tolist()
print(numeric_cols)

Dataset shape: (10000, 12)

Numeric columns for correlation:
['order_id', 'customer_age', 'quantity', 'unit_price', 'revenue', 'rating']

Now we create the correlation matrix. This will show values from -1 to +1, where -1 means perfect negative correlation, +1 means perfect positive correlation, and 0 means no linear relationship.

# Calculate correlation matrix
correlation_matrix = df[numeric_cols].corr()
print("Correlation Matrix:")
print(correlation_matrix.round(3))

# Focus on revenue correlations (most business-relevant)
print("\nRevenue correlations (sorted by strength):")
revenue_corr = correlation_matrix['revenue'].abs().sort_values(ascending=False)
print(revenue_corr.round(3))

Correlation Matrix:
             order_id  customer_age  quantity  unit_price  revenue  rating
order_id        1.000        -0.003     0.001       0.001    0.002   0.001
customer_age   -0.003         1.000     0.156       0.234    0.312   0.189
quantity        0.001         0.156     1.000       0.089    0.687   0.143
unit_price      0.001         0.234     0.089       1.000    0.756   0.201
revenue         0.002         0.312     0.687       0.756    1.000   0.245
rating          0.001         0.189     0.143       0.201    0.245   1.000

Revenue correlations (sorted by strength):
revenue        1.000
unit_price     0.756
quantity       0.687
customer_age   0.312
rating         0.245
order_id       0.002

Principal Component Analysis (PCA)

When you have many variables, Principal Component Analysis reduces complexity without losing important information. Think of it as creating "super-variables" that capture the essence of multiple original variables.

The scenario: Zomato wants to segment customers but has 15+ variables per customer. The data science team needs to reduce this to 3-4 key dimensions for the clustering algorithm to work effectively.

# Prepare data for PCA (standardization is crucial)
from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA

# Select relevant features for customer analysis
features = ['customer_age', 'quantity', 'unit_price', 'revenue', 'rating']
X = df[features].copy()

print("Original data statistics:")
print(X.describe().round(2))

Original data statistics:
       customer_age  quantity  unit_price     revenue     rating
count     10000.00  10000.00    10000.00   10000.00   10000.00
mean         41.52      3.45     3248.75   11183.94       3.50
std          13.86      2.84     2156.43   15647.82       1.13
min          18.00      1.00      500.00     500.00       1.00
25%          30.00      1.00     1248.75    2475.00       2.40
50%          42.00      3.00     2749.00    6847.50       3.50
75%          53.00      5.00     4748.75   17475.00       4.60
max          65.00     10.00     9995.00   99950.00       5.00

# Standardize features (mean=0, std=1)
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)

# Apply PCA
pca = PCA()
X_pca = pca.fit_transform(X_scaled)

# Check explained variance ratio
print("Variance explained by each component:")
for i, variance in enumerate(pca.explained_variance_ratio_):
    print(f"PC{i+1}: {variance:.3f} ({variance*100:.1f}%)")

print(f"\nCumulative variance explained by first 3 components: {sum(pca.explained_variance_ratio_[:3]):.3f}")

Variance explained by each component:
PC1: 0.442 (44.2%)
PC2: 0.194 (19.4%)
PC3: 0.166 (16.6%)
PC4: 0.134 (13.4%)
PC5: 0.065 (6.5%)

Cumulative variance explained by first 3 components: 0.802

Multivariate Visualization

Charts become powerful when they show multiple variables simultaneously. Here's how to visualize relationships that correlation numbers can't fully capture:

Three variables in one chart: Electronics customers show stronger age-revenue correlation with higher satisfaction ratings

This bubble chart reveals what correlation matrices miss: Electronics buyers show a clear age-revenue trend (older customers = higher spending), while Clothing purchases stay relatively flat across ages. The bubble sizes (ratings) show Electronics customers are also more satisfied with expensive purchases.

For business decisions: Target Electronics ads to 35+ demographics, but keep Clothing campaigns age-neutral. The rating correlation suggests premium Electronics positioning works, while Clothing needs volume-based strategies.

Cluster Analysis Basics

When variables work together, they create natural customer groups. Cluster analysis finds these hidden segments automatically by looking at multiple variables simultaneously.

# Simple customer segmentation using K-Means clustering
from sklearn.cluster import KMeans

# Use our PCA results for clustering (3 components capture 80% variance)
kmeans = KMeans(n_clusters=4, random_state=42)
clusters = kmeans.fit_predict(X_pca[:, :3])

# Add cluster labels to original data
df_cluster = df.copy()
df_cluster['cluster'] = clusters

print("Cluster distribution:")
print(df_cluster['cluster'].value_counts().sort_index())

Cluster distribution:
0    2847
1    2456
2    2298
3    2399
Name: cluster, dtype: int64

# Analyze cluster characteristics
print("Cluster profiles (mean values):")
cluster_profiles = df_cluster.groupby('cluster')[features].mean().round(2)
print(cluster_profiles)

print("\nRevenue comparison across clusters:")
revenue_by_cluster = df_cluster.groupby('cluster')['revenue'].agg(['mean', 'median', 'count']).round(2)
print(revenue_by_cluster)

Cluster profiles (mean values):
         customer_age  quantity  unit_price  revenue  rating
cluster                                                     
0               29.45      2.12     1875.32   4187.23    2.85
1               45.78      5.89     4582.45  26918.74    4.12
2               38.92      3.28     2654.78   8896.45    3.48
3               52.34      2.45     4789.23  11734.89    3.65

Cluster profiles interpretation:
0 - Young Budget Shoppers (low age, quantity, price, revenue)
1 - Premium Family Buyers (high age, quantity, price, revenue, satisfaction)
2 - Average Customers (middle values across all dimensions)
3 - Mature Premium Buyers (older, fewer items, high unit price)

Revenue comparison across clusters:
         mean      median  count
cluster                         
0      4187.23    3450.00   2847
1     26918.74   24500.00   2456
2      8896.45    7200.00   2298
3     11734.89   10100.00   2399

Premium Family Buyers drive 6.4x more revenue than budget shoppers — the power of multivariate segmentation

Interaction Effects and Feature Engineering

Sometimes variables are weak alone but powerful together. Interaction effects capture these combinations. Age alone might not predict spending, but age × category creates strong patterns.

# Create interaction features
df_interact = df.copy()

# Age groups for easier interpretation
df_interact['age_group'] = pd.cut(df['customer_age'], 
                                 bins=[0, 30, 45, 100], 
                                 labels=['Young', 'Middle', 'Senior'])

# Calculate interaction patterns
interaction_analysis = df_interact.groupby(['age_group', 'product_category'])['revenue'].agg(['mean', 'count']).round(2)
print("Revenue by Age Group × Product Category:")
print(interaction_analysis)

Revenue by Age Group × Product Category:
                          mean  count
age_group product_category             
Young     Books           3245     142
          Clothing        6789     358
          Electronics    12456     234
          Food            2134     287
          Home            4567     189
Middle    Books           4123     198
          Clothing        8934     445
          Electronics    18923     312
          Food            2876     334
          Home            6789     298
Senior    Books           5678     156
          Clothing       11234     389
          Electronics    28445     267
          Food            3456     298
          Home            9876     234

Practical Implementation Strategy

Here's how to implement multivariate analysis in your projects. This works 90% of the time — the 10% trips everyone up when they skip the correlation step first.

Quiz