Chi-Square (χ²)

Best Uses Case

Fraud Detection, Customer Segmentation

Features Types

Categorical

Target Type

Categorical

Chi-Square Test (χ²) for Feature Selection - Complete Guide

📋 Table of Contents

What is Chi-Square Test?

The Chi-Square test measures the independence between categorical variables. In feature selection, it tests whether a categorical feature and the target variable are independent or related.

High χ² score → Strong dependency → Feature is important
Low χ² score → Weak/no dependency → Feature is less important

When to Use It

✅ Use Chi-Square When:

Both feature AND target are CATEGORICAL (or binned numerical data)
You have discrete/nominal variables (e.g., color, gender, category)
You want to rank features by their relationship with the target
Working with classification problems

❌ Don't Use Chi-Square When:

Features are continuous numerical (use ANOVA F-test instead)
Target is continuous (use correlation or mutual information)
You have negative values (Chi-Square requires non-negative integers)
Sample size is very small (< 5 observations per cell)

Mathematical Intuition

Formula:

\chi^2 = \sum \frac{(O - E)^2}{E}

Observed: Actual frequency count in your data

Expected: Frequency if features were independent

P-value:

p-value < 0.05: Feature is significantly related to target (reject independence hypothesis)
p-value ≥ 0.05: No significant relationship (feature may be irrelevant)

Sklearn Implementation

Main Functions

1. `chi2()` - Calculate Chi-Square Statistics

from sklearn.feature_selection import chi2

chi_scores, p_values = chi2(X, y)

Parameters:

X: Feature matrix (2D array, non-negative values)
y: Target vector (1D array, categorical)

Returns:

chi_scores: Chi-square statistic for each feature
p_values: P-values for each feature

2. `SelectKBest()` - Select Top K Features

from sklearn.feature_selection import SelectKBest, chi2

selector = SelectKBest(score_func=chi2, k=5)
X_selected = selector.fit_transform(X, y)

Parameters:

score_func: Scoring function (use chi2)
k: Number of top features to select (integer or 'all')

Methods:

.fit_transform(X, y): Fit and transform data
.get_support(): Get boolean mask of selected features
.scores_: Chi-square scores
.pvalues_: P-values

3. `SelectPercentile()` - Select Top X% Features

from sklearn.feature_selection import SelectPercentile, chi2

selector = SelectPercentile(score_func=chi2, percentile=50)
X_selected = selector.fit_transform(X, y)

Parameters:

percentile: Percentage of features to keep (0-100)

Step-by-Step Workflow

Step 1: Import Libraries

import pandas as pd
import numpy as np
from sklearn.feature_selection import chi2, SelectKBest
from sklearn.datasets import load_breast_cancer
import matplotlib.pyplot as plt
import seaborn as sns

Step 2: Load and Prepare Data

# Load dataset
data = load_breast_cancer()
X = pd.DataFrame(data.data, columns=data.feature_names)
y = data.target

# For Chi-Square: Convert continuous to categorical (binning)
from sklearn.preprocessing import KBinsDiscretizer

discretizer = KBinsDiscretizer(n_bins=5, encode='ordinal', strategy='quantile')
X_categorical = discretizer.fit_transform(X)

Step 3: Calculate Chi-Square Scores

# Calculate chi-square statistics
chi_scores, p_values = chi2(X_categorical, y)

# Create results dataframe
results = pd.DataFrame({
    'Feature': X.columns,
    'Chi2_Score': chi_scores,
    'P_Value': p_values
}).sort_values('Chi2_Score', ascending=False)

print(results)

Step 4: Visualize Results

# Plot top 10 features
top_features = results.head(10)

plt.figure(figsize=(10, 6))
plt.barh(top_features['Feature'], top_features['Chi2_Score'])
plt.xlabel('Chi-Square Score')
plt.title('Top 10 Features by Chi-Square Test')
plt.gca().invert_yaxis()
plt.tight_layout()
plt.show()

Step 5: Select Features

# Method 1: Select top K features
selector = SelectKBest(chi2, k=10)
X_selected = selector.fit_transform(X_categorical, y)

# Get selected feature names
selected_features = X.columns[selector.get_support()].tolist()
print(f"Selected features: {selected_features}")

Step 6: Evaluate with ML Model

from sklearn.model_selection import cross_val_score
from sklearn.ensemble import RandomForestClassifier

# Compare models
model = RandomForestClassifier(random_state=42)

# Before feature selection
score_before = cross_val_score(model, X_categorical, y, cv=5).mean()

# After feature selection
score_after = cross_val_score(model, X_selected, y, cv=5).mean()

print(f"Accuracy before selection: {score_before:.4f}")
print(f"Accuracy after selection: {score_after:.4f}")

Practical Examples

Example 1: Binary Classification (Breast Cancer)

from sklearn.datasets import load_breast_cancer
from sklearn.feature_selection import chi2, SelectKBest
from sklearn.preprocessing import KBinsDiscretizer
import pandas as pd

# Load data
data = load_breast_cancer()
X = pd.DataFrame(data.data, columns=data.feature_names)
y = data.target

# Discretize features (required for chi-square)
discretizer = KBinsDiscretizer(n_bins=3, encode='ordinal', strategy='quantile')
X_discrete = discretizer.fit_transform(X)

# Apply chi-square test
chi_scores, p_values = chi2(X_discrete, y)

# Create results
results = pd.DataFrame({
    'Feature': X.columns,
    'Chi2_Score': chi_scores,
    'P_Value': p_values,
    'Significant': p_values < 0.05
}).sort_values('Chi2_Score', ascending=False)

print(results.head(10))

# Select top 5 features
selector = SelectKBest(chi2, k=5)
X_new = selector.fit_transform(X_discrete, y)
print(f"\nOriginal shape: {X_discrete.shape}")
print(f"Selected shape: {X_new.shape}")

Example 2: Multi-class Classification (Iris)

from sklearn.datasets import load_iris

# Load data
iris = load_iris()
X = pd.DataFrame(iris.data, columns=iris.feature_names)
y = iris.target

# Discretize
discretizer = KBinsDiscretizer(n_bins=4, encode='ordinal', strategy='uniform')
X_discrete = discretizer.fit_transform(X)

# Chi-square test
chi_scores, p_values = chi2(X_discrete, y)

results = pd.DataFrame({
    'Feature': X.columns,
    'Chi2_Score': chi_scores,
    'P_Value': p_values
}).sort_values('Chi2_Score', ascending=False)

print(results)

Example 3: Real Categorical Data

# Create sample categorical dataset
data = {
    'Color': ['Red', 'Blue', 'Green', 'Red', 'Blue', 'Green', 'Red', 'Blue'],
    'Size': ['S', 'M', 'L', 'M', 'L', 'S', 'L', 'M'],
    'Material': ['A', 'B', 'A', 'B', 'A', 'B', 'A', 'B'],
    'Purchased': [1, 0, 1, 1, 0, 1, 1, 0]
}

df = pd.DataFrame(data)

# Encode categorical variables
from sklearn.preprocessing import LabelEncoder

le = LabelEncoder()
X = df[['Color', 'Size', 'Material']].apply(le.fit_transform)
y = df['Purchased']

# Chi-square test
chi_scores, p_values = chi2(X, y)

print(pd.DataFrame({
    'Feature': X.columns,
    'Chi2_Score': chi_scores,
    'P_Value': p_values
}))

Common Pitfalls

1. Using Negative Values

# ❌ WRONG - Chi-square requires non-negative values
X_with_negatives = np.array([[-1, 2], [3, -4]])
# This will raise an error!

# ✅ CORRECT - Ensure all values are non-negative
from sklearn.preprocessing import MinMaxScaler
scaler = MinMaxScaler()
X_positive = scaler.fit_transform(X_with_negatives)

2. Not Discretizing Continuous Features

# ❌ WRONG - Continuous features directly
chi2(X_continuous, y)  # Results may be misleading

# ✅ CORRECT - Discretize first
from sklearn.preprocessing import KBinsDiscretizer
discretizer = KBinsDiscretizer(n_bins=5, encode='ordinal')
X_discrete = discretizer.fit_transform(X_continuous)
chi2(X_discrete, y)

3. Ignoring P-values

# Don't just look at chi-square scores
# Also check p-values for statistical significance

results = pd.DataFrame({
    'Feature': feature_names,
    'Chi2': chi_scores,
    'P_Value': p_values,
    'Significant': p_values < 0.05  # ← Important!
})

4. Small Sample Sizes

# Chi-square assumes each cell has at least 5 observations
# Check your contingency table

from scipy.stats import contingency

for col in X.columns:
    table = pd.crosstab(X[col], y)
    print(f"\n{col}:")
    print(table)
    # Ensure no cell has count < 5

Quick Reference Card

Task	Code
Basic chi-square	`chi2(X, y)`
Select top K	`SelectKBest(chi2, k=10).fit_transform(X, y)`
Select top 20%	`SelectPercentile(chi2, percentile=20).fit_transform(X, y)`
Get selected indices	`selector.get_support(indices=True)`
Get scores	`selector.scores_`
Get p-values	`selector.pvalues_`

🎯 Key Takeaways

Chi-square is for categorical data only
All values must be non-negative
Always check p-values (< 0.05 = significant)
Discretize continuous features before applying
Interpret results carefully - high score ≠ causation
Validate with ML models to confirm feature importance

Next Steps: Try ANOVA F-test for continuous features, or Mutual Information for mixed data types!

Chi-Square (χ²)

Chi-Square Test (χ²) for Feature Selection - Complete Guide

📋 Table of Contents

What is Chi-Square Test?

When to Use It

✅ Use Chi-Square When:

❌ Don't Use Chi-Square When:

Mathematical Intuition

Formula:

P-value:

Sklearn Implementation

Main Functions

1. chi2() - Calculate Chi-Square Statistics

2. SelectKBest() - Select Top K Features

3. SelectPercentile() - Select Top X% Features

Step-by-Step Workflow

Step 1: Import Libraries

Step 2: Load and Prepare Data

Step 3: Calculate Chi-Square Scores

Step 4: Visualize Results

Step 5: Select Features

Step 6: Evaluate with ML Model

Practical Examples

Example 1: Binary Classification (Breast Cancer)

Example 2: Multi-class Classification (Iris)

Example 3: Real Categorical Data

Common Pitfalls

1. Using Negative Values

2. Not Discretizing Continuous Features

3. Ignoring P-values

4. Small Sample Sizes

Quick Reference Card

🎯 Key Takeaways

1. `chi2()` - Calculate Chi-Square Statistics

2. `SelectKBest()` - Select Top K Features

3. `SelectPercentile()` - Select Top X% Features