Principal Component Analysis#

Principal Component Analysis or PCA is a dimensionality reduction technique that aims at reducing the number of extraneous variables to a smaller set of most important variables.

Problems with high dimensional data:

  • high computational cost while fitting a model

  • harder to visualise and analyse the dataset

  • may lead to overfitting and poor performance on unseen data

When to use PCA#

  • When processing data with multi-colinearity among the features/variables.

  • To find patterns in high-dimensional dataset

  • When it is difficult to identify variables to be completely removed from the training

  • To identify the directions in which the data is dispersed

Visualise a dataset using PCA Plot#

import matplotlib
from sklearn.datasets import load_iris as load_data
from sklearn.model_selection import train_test_split

matplotlib.rcParams["figure.figsize"] = (7, 7)
matplotlib.rcParams["font.size"] = 18

X, y = load_data(return_X_y=True)

from sklearn_evaluation import plot

plot.pca(X, y, target_names=["Setosa", "Versicolor", "Virginica"], n_components=3)

[<Axes: title={'center': 'Principal Component Plot'}, xlabel='Principal Component 1', ylabel='Principal Component 2'>,
 <Axes: title={'center': 'Principal Component Plot'}, xlabel='Principal Component 1', ylabel='Principal Component 3'>,
 <Axes: title={'center': 'Principal Component Plot'}, xlabel='Principal Component 2', ylabel='Principal Component 3'>]
../_images/ded69b7fbc7cda9213edc0b0294d57ce16d5a0be0fc3dc526e4571c6a8877f07.png ../_images/98726ea4666c4eca8ab0e21a2d986b2727920a22fcd11c02acde74497fc28d6e.png ../_images/2c8034d152e849ef16b97b0e4d6606c398e123da4f59a2b4976ffcc2812b96a5.png

Interpreting PCA plots#

  • PCA plots can help to reveal clusters. Data points that have similar features are clustered together.

  • First principal component captures the most variation in the data, while the second principal component reveals the second most variance.

Model fitting with PCA#

from sklearn.datasets import make_classification

X, y = make_classification(10000, n_features=5, n_informative=3, class_sep=0.5)

from sklearn.neighbors import KNeighborsClassifier

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.3, random_state=1)
knn = KNeighborsClassifier(n_neighbors=7)
knn.fit(X_train, y_train)
print("Accuracy before PCA:", knn.score(X_test, y_test))

Accuracy before PCA: 0.8973333333333333

Now let’s apply PCA on the data and retrain a model. We can see that the results with PCA are as good as without PCA. Accuracy is similar with lesser dimensions.

from sklearn.decomposition import PCA

pca = PCA(n_components=3)
X_new = pca.fit_transform(X)
X_train_new, X_test_new, y_train, y_test = train_test_split(
    X_new, y, test_size=0.3, random_state=1
)

knn = KNeighborsClassifier(n_neighbors=7)
knn.fit(X_train_new, y_train)
print("Accuracy after PCA:", knn.score(X_test_new, y_test))

Accuracy after PCA: 0.8973333333333333