How to Use PCA for Dimensionality Reduction in Python with sklearn

Use sklearn.decomposition.PCA to reduce data dimensions by fitting it to your data and transforming it. Initialize PCA with the number of components you want, then call fit_transform() on your dataset to get the reduced features.

Syntax

The basic syntax to use PCA in Python with sklearn is:

• PCA(n_components): Create a PCA object specifying how many dimensions to keep.
• fit_transform(data): Fit the PCA model to your data and reduce its dimensions in one step.
• explained_variance_ratio_: Attribute showing how much variance each principal component explains.

python

from sklearn.decomposition import PCA

pca = PCA(n_components=2)  # keep 2 principal components
reduced_data = pca.fit_transform(data)  # fit and reduce data dimensions
print(pca.explained_variance_ratio_)  # variance explained by each component

Example

This example shows how to reduce a 4-dimensional dataset to 2 dimensions using PCA and prints the reduced data and variance explained.

python

from sklearn.decomposition import PCA
from sklearn.datasets import load_iris

# Load sample data
iris = load_iris()
data = iris.data

# Initialize PCA to reduce to 2 components
pca = PCA(n_components=2)

# Fit PCA and transform data
reduced_data = pca.fit_transform(data)

# Print reduced data shape and first 5 rows
print('Reduced data shape:', reduced_data.shape)
print('First 5 rows of reduced data:\n', reduced_data[:5])

# Print variance explained by each component
print('Explained variance ratio:', pca.explained_variance_ratio_)

Output

Reduced data shape: (150, 2)
First 5 rows of reduced data:
 [[-2.68412563  0.31939725]
 [-2.71414169 -0.17700123]
 [-2.88899057 -0.14494943]
 [-2.74534286 -0.31829898]
 [-2.72871654  0.32675451]]
Explained variance ratio: [0.92461872 0.05306648]

Common Pitfalls

Common mistakes when using PCA include:

• Not scaling data before PCA, which can cause features with larger scales to dominate.
• Choosing too many or too few components without checking explained variance.
• Using PCA on categorical data without encoding.

Always scale your data (e.g., with StandardScaler) before applying PCA for best results.

python

from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA
from sklearn.datasets import load_iris

# Load data
iris = load_iris()
data = iris.data

# Wrong: PCA without scaling
pca_wrong = PCA(n_components=2)
reduced_wrong = pca_wrong.fit_transform(data)
print('Explained variance without scaling:', pca_wrong.explained_variance_ratio_)

# Right: Scale data first
scaler = StandardScaler()
data_scaled = scaler.fit_transform(data)
pca_right = PCA(n_components=2)
reduced_right = pca_right.fit_transform(data_scaled)
print('Explained variance with scaling:', pca_right.explained_variance_ratio_)

Output

Explained variance without scaling: [0.92461872 0.05306648]
Explained variance with scaling: [0.72962445 0.22850762]

Quick Reference

Remember these tips when using PCA:

• Use n_components to set how many dimensions to keep.
• Always scale your data before PCA.
• Check explained_variance_ratio_ to understand how much information is retained.
• Use fit_transform() to apply PCA in one step.

Key Takeaways

Initialize PCA with the number of components to keep using PCA(n_components).

Always scale your data before applying PCA to get meaningful results.

Use fit_transform() to reduce data dimensions in one step.

Check explained_variance_ratio_ to see how much variance each component captures.

PCA works best on numerical data and reduces redundancy by combining correlated features.