rnew icon6Grab Deal : Flat 30% off on live classes + 2 free self-paced courses! - SCHEDULE CALL rnew icon7


What Is K Medoids Clustering In Data Science?


Data science is a field that involves the use of various techniques to extract insights from data. One such technique is clustering, which involves grouping similar data points together based on some criteria. There are several types of clustering algorithms available in data science, and one popular algorithm is k medoids clustering.K medoids clustering, also known as partitioning around medoids (PAM), is an unsupervised learning algorithm used for cluster analysis. It is an extension of the k-means algorithm and has been widely used in different fields such as biology, finance, marketing, and social sciences.

In this blog post, we will explore what k medoids clustering is all about and how it differs from other clustering algorithms like k means. We will also look at how to implement k medoids using Python's scikit-learn library. For an in-depth understanding of the k-medoids clustering algorithm, our Data scientist course online helps you explore more about k medoids clustering, the most effective tool of data science.

What Is K Medoids Clustering?

K medoids clustering is a popular data mining and machine learning technique. This algorithm is particularly useful when dealing with large datasets, as it can efficiently group similar data points into clusters.One of the primary advantages of K medoids clustering over other methods is its ability to handle noisy or incomplete data. Since the representative objects (medoids) are actual observations themselves, they are less likely to be affected by outliers or missing values that could skew results in centroid-based approaches like K-means.Let's consider an example to understand better how K medoids clustering works. Suppose you have a dataset containing customer information for an online retailer. Each row represents a different customer, and each column contains information such as age, gender, location, purchase history, etc.Using K medoids clustering on this dataset would involve selecting k number of initial cluster centers (medoids), where k represents the desired number of clusters. These centers can be chosen randomly or using a hierarchical clustering method.Next, each observation in the dataset is assigned to its closest cluster center based on some distance metric (e.g., Euclidean distance). The distances between each observation and its corresponding cluster center are then calculated and summed up for all observations within that cluster.

The next step involves swapping out one of the existing medoids with another observation from that same cluster and recalculating the sum of distances for all observations within that new configuration. This process continues until no further improvements can be made by swapping out any more medoids.We have our final set of k clusters with their respective representative objects (medoids). These clusters can now be analyzed further to gain insights into customer behavior patterns or other trends within our data.In summary, K medoid clustering offers several advantages over traditional centroid-based methods like K-means due to its ability to handle noisy or incomplete data while producing more accurate overall results. By restricting representatives only to actual observations themselves, K medoids clustering provides a more robust and reliable approach to clustering analysis.

How Does K Medoids Clustering Work?

K medoids clustering is a popular and effective data clustering method used in various applications such as image segmentation, customer segmentation, and anomaly detection. The main difference between K medoids clustering and other types of clustering algorithms like K-means or hierarchical clustering is that it uses actual data points (called medoids) to represent each cluster instead of calculating the mean or centroid.

To understand how this algorithm works, let's consider an example where we have a dataset with 1000 observations on two variables: age and income. Our task is to group these observations into 'k' clusters based on their similarities.

Step 1: Selecting Initial Medoid Points

We start by selecting 'k' number of random data points from our dataset as our initial medoid points. Let's say we choose k=3 for this example, so we randomly select three data points from our dataset as our initial medoids.

Step 2: Assigning Non-Medoid Points to The Nearest Medoid

Next, we assign each non-medoid point in the dataset to its nearest medoid based on some distance metric (e.g., Euclidean distance). This means that every observation in the dataset will be assigned to one of the three selected initial centroids based on which centroid they are closest to.

Step 3: Calculating Total Cost

After assigning all non-medoid points to their respective centroids, we calculate the total cost function for each cluster using some distance measure (e.g., summing up distances between all non-medoids assigned within the same centroid). This cost function represents how well all data points within the same centroid are clustered together.

Step 4: Swapping Non-Mediod Points with Current Selected Mediod 

Now comes an iterative process where we try swapping one non-medoid point at a time with one current selected mediod at a time until there's no further decrease in cost function i.e. when the overall cost function is minimized. This process helps us find better medoids within each cluster, improving our clustering accuracy.

Step 5: Repeat Until Convergence

Finally, we repeat steps 2-4 until convergence i.e., when there's no further improvement in the cost function or changes in clusters occur.

In summary, K medoids clustering works by selecting initial random data points as medoids and then iteratively assigning non-medoid points to their nearest centroids while trying out different combinations of swapping selected medoids with a non-medoid point to minimize the overall cost. By following these steps, we can group similar observations into the 'k' number of clusters based on their similarities and differences.

K-Means VS K-Medoids

K-means clustering is a commonly used algorithm for partitioning data into k clusters based on their similarity. The algorithm works by randomly selecting k centroids and then assigning each point in the dataset to its nearest centroid. The centroid is then updated as the mean of all points assigned to it, and this process continues until convergence.One limitation of k-means clustering is that it assumes that the centroid of each cluster can be any point within the cluster. This means that if there are outliers in a particular cluster, they could affect the position of the centroid leading to suboptimal results.

K-medoid clustering, also known as Partitioning Around Medoids (PAM), overcomes this limitation by using actual observations themselves instead of centroids as representatives or medoids for each cluster. In other words, rather than computing an average value for all points assigned to a given cluster, like in K-means clustering, we select one observation from our dataset that best represents that group.This approach makes K-medoid more robust against noise and outliers since it selects only actual observations as representatives; therefore, outliers do not have much effect on them. Additionally, unlike K means, where the distance between data points is usually measured using Euclidean distance metric, which may not always work well when dealing with categorical variables or non-linear relationships between variables, K medoid allows us to use different dissimilarity measures such as Manhattan distance or cosine similarity depending on our problem requirements.

For instance, consider a scenario where we want to segment customers based on their purchase history and demographic information such as age and gender. Suppose some customers have significantly higher spending habits than others due to exceptional circumstances like buying gifts during holiday seasons or making bulk orders occasionally. In that case, these high spenders will appear far away from other customers, thus affecting how clusters are formed under the k-means algorithm. Still, we can expect better clustering results, thanks to K-medoids' ability to pick representative medoids from existing data points without considering outlier data points.

While K-means is a widely used algorithm for cluster analysis, it has some limitations that can be overcome using K-medoid. The use of actual observations as representatives rather than centroids makes K-medoid more robust against noise and outliers. Moreover, the ability to choose different dissimilarity measures in K-medoid allows us to tailor our approach based on specific requirements or dataset characteristics.

Implementing K Medoids Clustering Using Python's Scikit-Learn Library

Python provides various libraries for machine learning tasks such as scikit-learn, including several modules for different unsupervised learning algorithms, including K-Medoids.

To implement K-Medoids using Python's scikit-learn library, we need first to install this package by running !pip install -U scikit-learn command on your terminal or Jupyter notebook environment, then import the necessary modules:


from sklearn_extra.cluster import KMedoids

from sklearn.datasets import make_blobs

import matplotlib.pyplot as plt


After importing the necessary packages and generating data points with the make_blobs function from sklearn.datasets module, we can create our model instance by specifying number clusters (k) and metric distance measure:


model = KMedoids(n_clusters=3,distance_metric='euclidean')


y_pred = model.predict(X)


Finally plotting our result using Matplotlib library:


plt.scatter(X[:,0], X[:,1], c=y_pred)



Partitioning Methods In Large Databases: From K-Medoids To Clarans

Although PAM algorithms easily employ k-medoids to partition data on smaller sets, they struggle when it comes to scaling up. However, a sampling-based method called CLARA (Clustering LARge Applications) can handle larger volumes of data.

CLARA works by selecting a subset of data for analysis instead of studying the entire collection. The basic concept involves randomly sampling the dataset and using PAM algorithms to choose medoids from this pool. This way, selected items will likely reflect what would have been chosen if we had studied all objects in the dataset.

We can return with an optimal clustering solution by repeating this process several times and applying PAM algorithms each time. It's no surprise that CLARA performs better than PAM when dealing with huge amounts of data since its complexity per iteration is O(ks2 +k(nk)), where s represents sample size, and k denotes number clusters. At the same time, n refers to the total objects in our database.

Based on experimental evidence, it has been determined that CLARANS outperforms both PAM and CLARA. By utilizing a silhouette coefficient to gauge an object's degree of belonging to a cluster, the most "natural" number of clusters can be identified. Additionally, CLARANS is capable of identifying anomalies. However, it should be noted that its computational complexity is approximately O(n 2 ), where n represents the total number of items, and the quality of its grouping may be influenced by the sampling technique employed. To improve CLARANS' ability to handle data objects on disc, incorporating spatial data structures like R*-trees would prove beneficial.

Advantages of Using K Medoids Clustering

K-medoids have various advantages of using clustering algorithms in data science. Here are a few of the advantages of using the k-medoids clustering algorithm:

  1. Robustness: As mentioned earlier, k-medoids are more robust to outliers than k-means.
  2. Interpretable Results: Since the medoid points are actual data points from the dataset, it's easier to interpret and explain results compared to other clustering algorithms.
  3. Scalability: It works well with large datasets since it only needs pairwise distances between objects instead of computing all pairwise similarities like hierarchical clustering.

Disadvantages of Using K Medoids Clustering

Despite having huge advantages of using k medoids, it also comes with a few limitations mentioned below:

  1. Computationally Expensive: The algorithm has a higher computational cost compared to k-means due to its swapping step.
  2. Sensitivity To Initial Points Selection: The final result may vary depending on the initial medoid point selection. 
Application of K-Medoids Clustering 

Despite these limitations, K-medoid clustering has several real-world applications where it's commonly used:  

  • It's frequently used in healthcare for patient segmentation based on clinical parameters such as age, gender, medical history, etc., which can help hospitals tailor treatments more effectively for different groups of patients.
  • In finance industries like banking and insurance, companies use K-Mediods Clustering techniques for fraud detection purposes; they analyze transactions made with credit cards by grouping them into clusters according to similar patterns found across multiple accounts so potential fraudulent activity can be detected early before any major losses occur.
  • Another application includes customer segmentation in marketing, where businesses segment customers into groups based on demographics or purchasing behavior so they can tailor their marketing efforts more effectively.

cta10 icon

Data Science Training

  • Personalized Free Consultation
  • Access to Our Learning Management System
  • Access to Our Course Curriculum
  • Be a Part of Our Free Demo Class


K medoids clustering is a powerful unsupervised learning algorithm for cluster analysis. It differs from other clustering algorithms like k means in that it restricts representatives only to be actual observations themselves and handles outliers differently.

Python's scikit-learn library provides an easy way to implement k-medoid clustering, making it accessible to data scientists of all levels. By understanding the basics of k-medoid clustering and how to implement it using Python, you can start exploring your datasets and extracting valuable insights from them. You can also learn about neural network guides and python for data science if you are interested in further career prospects in data science.

Trending Courses

Cyber Security icon

Cyber Security

  • Introduction to cybersecurity
  • Cryptography and Secure Communication 
  • Cloud Computing Architectural Framework
  • Security Architectures and Models
Cyber Security icon1

Upcoming Class

16 days 05 Jul 2024

QA icon


  • Introduction and Software Testing
  • Software Test Life Cycle
  • Automation Testing and API Testing
  • Selenium framework development using Testing
QA icon1

Upcoming Class

8 days 27 Jun 2024

Salesforce icon


  • Salesforce Configuration Introduction
  • Security & Automation Process
  • Sales & Service Cloud
  • Apex Programming, SOQL & SOSL
Salesforce icon1

Upcoming Class

7 days 26 Jun 2024

Business Analyst icon

Business Analyst

  • BA & Stakeholders Overview
  • BPMN, Requirement Elicitation
  • BA Tools & Design Documents
  • Enterprise Analysis, Agile & Scrum
Business Analyst icon1

Upcoming Class

9 days 28 Jun 2024

MS SQL Server icon

MS SQL Server

  • Introduction & Database Query
  • Programming, Indexes & System Functions
  • SSIS Package Development Procedures
  • SSRS Report Design
MS SQL Server icon1

Upcoming Class

9 days 28 Jun 2024

Data Science icon

Data Science

  • Data Science Introduction
  • Hadoop and Spark Overview
  • Python & Intro to R Programming
  • Machine Learning
Data Science icon1

Upcoming Class

2 days 21 Jun 2024

DevOps icon


  • Intro to DevOps
  • GIT and Maven
  • Jenkins & Ansible
  • Docker and Cloud Computing
DevOps icon1

Upcoming Class

5 days 24 Jun 2024

Hadoop icon


  • Architecture, HDFS & MapReduce
  • Unix Shell & Apache Pig Installation
  • HIVE Installation & User-Defined Functions
  • SQOOP & Hbase Installation
Hadoop icon1

Upcoming Class

2 days 21 Jun 2024

Python icon


  • Features of Python
  • Python Editors and IDEs
  • Data types and Variables
  • Python File Operation
Python icon1

Upcoming Class

17 days 06 Jul 2024

Artificial Intelligence icon

Artificial Intelligence

  • Components of AI
  • Categories of Machine Learning
  • Recurrent Neural Networks
  • Recurrent Neural Networks
Artificial Intelligence icon1

Upcoming Class

10 days 29 Jun 2024

Machine Learning icon

Machine Learning

  • Introduction to Machine Learning & Python
  • Machine Learning: Supervised Learning
  • Machine Learning: Unsupervised Learning
Machine Learning icon1

Upcoming Class

23 days 12 Jul 2024

 Tableau icon


  • Introduction to Tableau Desktop
  • Data Transformation Methods
  • Configuring tableau server
  • Integration with R & Hadoop
 Tableau icon1

Upcoming Class

2 days 21 Jun 2024