Difference between revisions of "Python-for-Machine-Learning/C3/K-Means-Clustering/English"

• Introduction to K-means Clustering
• Steps involved in K-means Algorithm
• Choosing the Optimal number of Clusters (k)

System Requirements

• Ubuntu Linux OS version 24.04
• Jupyter Notebook IDE

• The Learner must have basic knowledge of Python
• For prerequisite Python tutorials, please visit this website

Code files

• The files used in this tutorial are provided in the Code files link.
• Please download and extract the files.
• Make a copy and then use them while practicing.

K-means Clustering

• K-means clustering is an unsupervised machine learning algorithm.
• It partitions data into a predetermined number of clusters.
• This algorithm aims to group similar data points together.

K-means Clustering

• It creates distinct clusters by separating dissimilar points.

Show Slide:

Working working.png

• First, choose the number of clusters, k, to divide the data.
• Then, randomly pick k data points as initial centroids which is the center of each cluster.
• It represents the average position of all points in that cluster.

Working working.png

• Each data point is assigned to the nearest cluster centroid.
• Update the centroids by averaging the data points in each cluster.

Working working.png

• New centroids replace the old ones based on these mean values.

• This is repeated until the cluster assignments no longer change.

Point to KMeansClustering.ipynb

K means clustering dot ipynb is the python notebook file used in this tutorial.

The goal is to group individuals based on income and spending patterns.

Type conda activate ml

Press Enter

Activate the machine learning environment as shown.

Type cd Downloads

Type jupyter notebook

Please navigate to the respective folder of your code file location.

Then type, jupyter space notebook and press Enter.

KMeansClustering.ipynb file

Click the K means clustering dot ipynb file to open it.

import numpy as np

import pandas as pd

Make sure to Press Shift and Enter to execute the code in each cell.

Let us display the first few rows.

print(df.describe())

Hover over the output

The dataset contains 200 entries for Annual Income and Spending Score.

Income ranges from 15K to 137K, while the spending score varies from 1 to 99.

df.isnull()

Hover over the output

The output shows False for all entries, confirming no missing values.

We also use boxplots to detect potential outliers.

Income is more spread out, while spending scores form distinct clusters.

The boxplot identifies an outlier in the Annual Income feature.

It is indicated by a point above the upper whisker.

The Spending Score feature does not exhibit clear outliers.

Highlight:

scaler = MinMaxScaler()

scaled_df = scaler.fit_transform(df)

First, we normalize the dataset using MinMaxScaler to scale feature values.

This transformation ensures all values fall between 0 and 1.

It improves uniformity and enhances the clustering performance.

silhouette_scores = []

K_range = range(2, 11)

for K in K_range:

kmeans = KMeans(n_clusters=K, random_state=42, n_init=10)

cluster_labels = kmeans.fit_predict(scaled_df)

score = silhouette_score(scaled_df, cluster_labels)

silhouette_scores.append(score) plt.show()

The highest Silhouette score suggests the best k value.

To do this, we initialize an empty list to store silhouette scores.

The for loop iterates through a range of K values to perform clustering.

For each K, a KMeans model is created and trained.

Cluster labels are predicted, and the silhouette score is calculated.

The scores are stored for further analysis.

Further, we generate a plot to visualize the silhouette scores.

Hover over output plot

The plot shows silhouette scores for different cluster values.

A higher score means better defined clusters with less overlap.

In this plot, the optimal k is 5 as it has the highest silhouette score.

optimal_k = 5

Highlight:

The model is fitted to the scaled data, assigning cluster labels.

The coordinates of these centers are printed for analysis.

Each row corresponds to a centroid’s position for income and spending score.

df['Cluster'] = kmeans.labels_

A new column Cluster is added to store cluster labels.

silhouette_avg = silhouette_score(scaled_df, kmeans.labels_)

print("Silhouette Score: ", silhouette_avg)

A higher score indicates well-separated and dense clusters.

A score close to -1 means the data point is poorly clustered.

A score close to 1 means the data point is well-clustered.

This suggests that while clusters are distinguishable, some overlap may exist.

plt.figure(figsize=(8, 5))

plt.show()

To enhance clarity, each point is color-coded based on its cluster assignment.

The cluster centers are marked with red X for easy identification.

This visualization helps to understand how data points are grouped.

Each dot represents a data point, color-coded by its assigned cluster.

The clusters show distinct groupings based on annual income and spending score.

Summary

Let us summarize.

Assignment

• Use Elbow method instead of Silhouette method to find the optimal k
• Use inertia to identify it
• Then, plot the elbow curve

Assignment

Show elbow.png

Assignment Solution

Show plot.png

FOSSEE Forum

Thank you

Thanks for joining.