Clustering | Different Methods, and Applications (Updated 2024)

Sauravkaushik8 Kaushik 13 Jun, 2024 • 9 min read

Introduction

When encountering an unsupervised learning problem initially, confusion may arise as you aren’t seeking specific insights but rather identifying data structures. This process, known as clustering or cluster analysis, identifies similar groups within a dataset.

It is one of the most popular clustering techniques in data science used by data scientists. Entities in each group are comparatively more similar to entities of that group than those of the other groups. In this article, I will be taking you through the types of clustering, different clustering algorithms, and a comparison between two of the most commonly used methods of clustering in machine learning.

Note: To learn more about clustering and other machine learning algorithms (both supervised and unsupervised) check out the following courses-

Learning Objectives

  • Learn about Clustering in machine learning, one of the most popular unsupervised classification techniques.
  • Get to know K means and hierarchical clustering and the difference between the two.

What Is Clustering in Machine Learning?

Clustering is the task of dividing the unlabeled data or data points into different clusters such that similar data points fall in the same cluster than those which differ from the others. In simple words, the aim of the clustering process is to segregate groups with similar traits and assign them into clusters.

Let’s understand this with an example. Suppose you are the head of a rental store and wish to understand the preferences of your customers to scale up your business. Is it possible for you to look at the details of each customer and devise a unique business strategy for each one of them? Definitely not. But, what you can do is cluster all of your customers into, say 10 groups based on their purchasing habits and use a separate strategy for customers in each of these 10 groups. And this is what we call clustering methods.

Now that we understand what clustering is. Let’s take a look at its different types.

Types of Clustering in Machine Learning

Clustering broadly divides into two subgroups:

  • Hard Clustering: Each input data point either fully belongs to a cluster or not. For instance, in the example above, every customer is assigned to one group out of the ten.
  • Soft Clustering: Rather than assigning each input data point to a distinct cluster, it assigns a probability or likelihood of the data point being in those clusters. For example, in the given scenario, each customer receives a probability of being in any of the ten retail store clusters.

Different Types of Clustering Algorithms

Since the task of clustering methods is subjective, the means that can be used for achieving this goal are plenty. Every methodology follows a different set of rules for defining the ‘similarity’ among data points. In fact, there are more than 100 clustering algorithms known. But few of the algorithms are used popularly. Let’s look at them in detail:

Connectivity Models

As the name suggests, these models are based on the notion that the data points closer in data space exhibit more similarity to each other than the data points lying farther away. These models can follow two approaches. In the first approach, they start by classifying all data points into separate clusters & then aggregating them as the distance decreases. In the second approach, all data points are classified as a single cluster and then partitioned as the distance increases. Also, the choice of distance function is subjective. These models are very easy to interpret but lack scalability for handling big datasets. Examples of these models are the hierarchical clustering algorithms and their variants.

Centroid Models

These clustering algorithms iterate, deriving similarity from the proximity of a data point to the centroid or cluster center. The k-Means clustering algorithm, a popular example, falls into this category. These models necessitate specifying the number of clusters beforehand, requiring prior knowledge of the dataset. They iteratively run to discover local optima.

Distribution Models

These clustering models are based on the notion of how probable it is that all data points in the cluster belong to the same distribution (For example: Normal, Gaussian). These models often suffer from overfitting. A popular example of these models is the Expectation-maximization algorithm which uses multivariate normal distributions.

Density Models

These models search the data space for areas of the varied density of data points in the data space. They isolate different dense regions and assign the data points within these regions to the same cluster. Popular examples of density models are DBSCAN and OPTICS. These models are particularly useful for identifying clusters of arbitrary shape and detecting outliers, as they can detect and separate points that are located in sparse regions of the data space, as well as points that belong to dense regions.

Now I will be taking you through two of the most popular clustering algorithms in detail – K Means and Hierarchical. Let’s begin.

K Means Clustering

K means is an iterative clustering algorithm that aims to find local maxima in each iteration. This algorithm works in these 5 steps:

Step1:

Specify the desired number of clusters K: Let us choose k=2 for these 5 data points in 2-D space.

clustering, k-means | Clustering in Machine Learning

Step 2:

Randomly assign each data point to a cluster: Let’s assign three points in cluster 1, shown using red color, and two points in cluster 2, shown using grey color.

k-means clustering

Step 3:

Compute cluster centroids: The centroid of data points in the red cluster is shown using the red cross, and those in the grey cluster using a grey cross.

k-means, centroid

Step 4:

Re-assign each point to the closest cluster centroid: Note that only the data point at the bottom is assigned to the red cluster, even though it’s closer to the centroid of the grey cluster. Thus, we assign that data point to the grey cluster.

centroid | Clustering in Machine Learning

Step 5:

Re-compute cluster centroids: Now, re-computing the centroids for both clusters.

clustering, centroid, k-means | Clustering in Machine Learning

Repeat steps 4 and 5 until no improvements are possible: Similarly, we’ll repeat the 4th and 5th steps until we’ll reach global optima, i.e., when there is no further switching of data points between two clusters for two successive repeats. It will mark the termination of the algorithm if not explicitly mentioned.

Here is a live coding window where you can try out K Means Algorithm using the scikit-learn library.

Hierarchical Clustering

Hierarchical clustering methods, as the name suggests, is an algorithm that builds a hierarchy of clusters. This algorithm starts with all the data points assigned to a cluster of their own. Then two nearest clusters are merged into the same cluster. In the end, this algorithm terminates when there is only a single cluster left.

The results of hierarchical clustering can be shown using a dendrogram. The dendrogram can be interpreted as:

hierarchical clustering, dendogram | Clustering in Machine Learning

At the bottom, we start with 25 data points, each assigned to separate clusters. The two closest clusters are then merged till we have just one cluster at the top. The height in the dendrogram at which two clusters are merged represents the distance between two clusters in the data space.

The decision of the no. of clusters that can best depict different groups can be chosen by observing the dendrogram. The best choice of the no. of clusters is the no. of vertical lines in the dendrogram cut by a horizontal line that can transverse the maximum distance vertically without intersecting a cluster.

In the above example, the best choice of no. of clusters will be 4 as the red horizontal line in the dendrogram below covers the maximum vertical distance AB.

hierarchical clustering | Clustering Methods

Important Points for Hierarchical Clustering

  • This algorithm has been implemented above using a bottom-up approach. It is also possible to follow a top-down approach starting with all data points assigned in the same cluster and recursively performing splits till each data point is assigned a separate cluster.
  • The decision to merge two clusters is taken on the basis of the closeness of these clusters. There are multiple metrics for deciding the closeness of two clusters:
    • Euclidean distance: ||a-b||2 = √(Σ(ai-bi))
    • Squared Euclidean distance: ||a-b||22 = Σ((ai-bi)2)
    • Manhattan distance: ||a-b||1 = Σ|ai-bi|
    • Maximum distance:||a-b||INFINITY = maxi|ai-bi|
    • Mahalanobis distance: √((a-b)T S-1 (-b))   {where, s : covariance matrix}

Difference Between K Means and Hierarchical Clustering

  • Hierarchical clustering methods can’t handle big data well, but K Means can. This is because the time complexity of K Means is linear, i.e., O(n), while that of hierarchical is quadratic, i.e., O(n2).
  • Since we start with a random choice of clusters, the results produced by running the algorithm multiple times might differ in K Means clustering. While in Hierarchical clustering, the results are reproducible.
  • K Means is found to work well when the shape of the clusters is hyperspherical (like a circle in 2D or a sphere in 3D).
  • K Means clustering requires prior knowledge of K, i.e., no. of clusters you want to divide your data into. But, you can stop at whatever number of clusters you find appropriate in hierarchical clustering by interpreting the dendrogram.

Applications of Clustering

Clustering has a large no. of application of clustering spread across various domains. Some of the most popular applications of clustering are recommendation engines, market segmentation, social network analysis, search result grouping, medical imaging, image segmentation, and anomaly detection.

Improving Supervised Learning Algorithms With Clustering

Clustering is an unsupervised machine learning approach, but can it be used to improve the accuracy of supervised machine learning algorithms as well by clustering the data points into similar groups and using these cluster labels as independent variables in the supervised machine learning algorithm? Let’s find out.

Let’s check out the impact of clustering on the accuracy of our model for the classification problem using 3000 observations with 100 predictors of stock data to predict whether the stock will go up or down using R. This dataset contains 100 independent variables from X1 to X100 representing the profile of a stock and one outcome variable Y with two levels: 1 for the rise in stock price and -1 for drop in stock price.

The dataset is available here: Download

Let’s first try applying random forest without clustering in python.

#loading required libraries
library('randomForest')

library('Metrics')

#set random seedset.seed(101)

#loading dataset

data<-read.csv("train.csv",stringsAsFactors= T)

#checking dimensions of datadim(data)

## [1] 3000  101

#specifying outcome variable as factor


 data$Y<-as.factor(data$Y)

#dividing the dataset into train and testtrain<-data[1:2000,]
test<-data[2001:3000,]

#applying randomForest model_rf<-randomForest(Y~.,data=train)

preds<-predict(object=model_rf,test[,-101])

table(preds)

## preds
##  -1   1
## 453 547

#checking accuracy

auc(preds,test$Y)

## [1] 0.4522703

So, the accuracy we get is 0.45. Now let’s create five clusters based on values of independent variables using k-means and reapply random forest.

#combing test and train

all<-rbind(train,test)

#creating 5 clusters using K- means clustering

Cluster <- kmeans(all[,-101], 5)

#adding clusters as independent variable to the dataset.all$cluster<-as.factor(Cluster$cluster)

#dividing the dataset into train and testtrain<-all[1:2000,]
test<-all[2001:3000,]

#applying randomforestmodel_rf<-randomForest(Y~.,data=train)

preds2<-predict(object=model_rf,test[,-101])

table(preds2)

## preds2

## -1   1 

##548 452 

auc(preds2,test$Y)

## [1] 0.5345908

Whoo! In the above example, even though the final accuracy is poor but clustering has given our model a significant boost from an accuracy of 0.45 to slightly above 0.53.

This shows that clustering can indeed be helpful for supervised machine-learning tasks.

Conclusion

In this article, we have discussed the various ways of performing clustering. We came across application of clustering for unsupervised learning in a large no. of domains and also saw how to improve the accuracy of a supervised machine learning algorithm using clustering.

Although clustering is easy to implement, you need to take care of some important aspects, like treating outliers in your data and making sure each cluster has a sufficient population. These aspects of clustering are dealt with in great detail in this article.

Key Takeaways

  • Clustering helps to identify patterns in data and is useful for exploratory data analysis, customer segmentation, anomaly detection, pattern recognition, and image segmentation.
  • It is a powerful tool for understanding data and can help to reveal insights that may not be apparent through other methods of analysis.
  • Its types include partition-based, hierarchical, density-based, and grid-based clustering.
  • The choice of clustering algorithm and the number of clusters to use depend on the nature of the data and the specific problem at hand.

Frequently Asked Questions

Q1. What is clustering in machine learning?

A. Clustering in machine learning involves grouping similar data points together based on their features, allowing for pattern discovery without predefined labels.

Q2. What is clustering and its type?

A. Clustering is a method of unsupervised learning where data points are grouped based on similarity. Types include K-means, hierarchical, DBSCAN, and mean shift.

Q3. What is an example of clustering?

A. An example of clustering is customer segmentation, where a business groups customers based on purchasing behavior to tailor marketing strategies.

Q4. How does clustering work?

A. Clustering works by evaluating the distances or similarities between data points, then grouping them into clusters where intra-cluster similarity is maximized and inter-cluster similarity is minimized.

Saurav is a Data Science enthusiast, currently in the final year of his graduation at MAIT, New Delhi. He loves to use machine learning and analytics to solve complex data problems.

Frequently Asked Questions

Lorem ipsum dolor sit amet, consectetur adipiscing elit,

Responses From Readers

Clear

Ankit Gupta
Ankit Gupta 03 Nov, 2016

Very nice tutorial Saurav!

Richard Warnung
Richard Warnung 03 Nov, 2016

Nice, post! Please correc the last link - it is broken - thanks!

Sai Satheesh G
Sai Satheesh G 03 Nov, 2016

I accept that clustering may help in improving the supervised models. But here in the above: Clustering is performed on sample points (4361 rows). Is that right.? But I think correct way is to cluster features (X1-X100) and to represent data using cluster representatives and then perform supervised learning. Can you please elaborate further? Why samples are being clustered in the code (not independent variables)?

Luis
Luis 03 Nov, 2016

Nice article! How would you handle a clustering problem when there are some variables with many missing values (let's say...around 90% of each column). These missing values are not random at all, but even they have a meaning, the clustering output yields some isolated (and very small) groups due to these missing values. Thanks in advance!

Kunal Dash
Kunal Dash 03 Nov, 2016

Hello Saurav, Your article and related explanation on clustering and the two most used methods was very insightful. However, please do enlighten us by telling how does one interpret cluster output for both these methods - K-means and Hierarchical. Also, it would be nice if you could let the reader know when could one use K-means versus say something like K-median. In what scenario does the former work and in which one does the latter??? It would also be a great idea to: 1. Discuss the ways to implement a density based algorithm and a distribution based one 2. Maybe show an actual example of market segmentation You have done a good job of showing how clustering could in sense preclude a following classification method but if the problem is such that it is only limited to clustering, then how would you explain the output to an uninitiated audience? Maybe some thoughts for your second article in the clustering article. But great job. I enjoyed reading your piece.

Kunal Dash
Kunal Dash 03 Nov, 2016

Also Saurav, It might be a good idea to suggest which clustering algorithm would be appropriate to use when: 1. All variables are continuous 2. All variables are categorical - many times this could be the case 3. All variables are count - maybe sometimes 4. A mix of continuous and categorical - this could be possibly the most common 5. Similarly a mix of continuous, categorical and count To be more precise, if I had one or more scenarios above, and was using a distance based method to calculate distances between points, what distance calculation method works where. Any insights would be great!!

Nikunj Agarwal
Nikunj Agarwal 04 Nov, 2016

Hi Saurav, Since we are classifying assets in this tutorial, don't you think corelation based distance should give us better results than eucledian distances (which k-means normally uses)?

Kern Paillard
Kern Paillard 07 Nov, 2016

Hi It 's a good post on covering a broad topic like Clustering. However, I'm not so convinced about using Clustering for aiding Supervised ML. For me, Clustering based approaches tend to be more 'exploratory' in nature to understand the inherent data structure, segments et al. Dimensionality Reduction techniques like PCA are more intuitive approaches (for me) in this case, quite simple because you don't get any dimensionality reduction by doing clustering and vice-versa, yo don't get any groupings out of PCA like techniques. my distinction of the two, PCA is used for dimensionality reduction / feature selection / representation learning e.g. when the feature space contains too many irrelevant or redundant features. The aim is to find the intrinsic dimensionality of the data. K-means is a clustering algorithm that returns the natural grouping of data points, based on their similarity. I'd like to point to the excellent explanation and distinction of the two on Quora : https://www.quora.com/What-is-the-difference-between-factor-and-cluster-analyses my question to you how would you fit / cluster the same groupings (you obtained out of clustering the training set) onto a unseen test set? or would you apply clustering to it again? typically, you perform PCA on a training set and apply the same loadings on to a new unseen test set and not fit a new PCA to it..

Aditya
Aditya 13 Nov, 2016

Really nice article Saurav , this helped me understand some of the basic concepts regarding clustering. I was hoping if you can post similar articles on Fuzzy, DBSCAN, Self Organizing Maps. Aditya

J Vitor da Silva
J Vitor da Silva 14 Nov, 2016

Hi Saurav Kaushik, I am new to this area, but I am in search of help to understand it deeper. One of my personal projects involves analysing data for creating a "predictive model" based on some information collected about previous historical data which I have in a spreadsheet (or in .txt file if it is bette). Could you recommend a simple package (in Python or in Delphi) that can help me do something like this? My spreadsheet has (for example), 1500 lines which represent historical moments (Test 1, Test2...Test1500). On the columns, I have the Labels and Values for each of 1000 characteristics I analyse separately at each Test. What I would like to do with this? To be able to "predict" some 10 ou 20 values for 10 or 20 characteristics for the next Test1501. Do you think it is possible? If you are involved in this kind of project, what would it cost me to have your help in building a tool for doing that? I can send you an example file, if you would be interested in helping me. My direct contact : dixiejoelottolex at gmail dot com

Laurent
Laurent 17 Nov, 2016

Hi and thank you for your article. Running your example I am running in a series of issues. The first one being the result of preds<-predict(object=model_rf,test[,-101]) head(table(preds)) preds -0.192066666666667 -0.162533333333333 -0.120533333333333 -0.0829333333333333 -0.0793333333333333 1 1 1 1 1 -0.079 1 Then auc(preds,test$Y) [1] NaN The second exemple with the added cluster produces the same result. Any idea why my result is so different than yours?

Shirish
Shirish 12 Dec, 2016

Hey Saurav, Could you please give a code for Python? Thanks

Hitesh Waghela
Hitesh Waghela 19 Jun, 2017

Hi Saurav, In the last section, where you calculate Area under ROC (auc), the syntax seems incorrect. The syntax mentioned for auc in help is : auc(actual , predicted) whereas you have taken auc(predicted, actual). Thanks, Hitesh

Lakshmi
Lakshmi 06 Jul, 2017

Hi Saurav I am new to this area and want to further look into it. Can you tell me how to include the package metaviz in R?

Tawfeek Shah
Tawfeek Shah 30 Jul, 2017

Thank you so much for such a beautiful lesson. This was the first time I was trying to understand Clustering and I have nailed it, Thanks to you!

NG
NG 01 Nov, 2017

Hi, can K means be applied on Geo coordinates I have a data on 100 locations (lat Long) which i have to group in 10 clusters and assign each cluster to a person and also identify a central location frm where this person operates. query 1 : can i use k means on geo coordinates to form clusters query 2: centroids are the means of the clusters, so these can be different than the actual observation values. how do i find out actual observation closest to centroid

Mohammed Abdul Raoof
Mohammed Abdul Raoof 05 Feb, 2018

Hi Saurav, It is Good for understanding but add the elbow method

Luca Scrucca
Luca Scrucca 28 Apr, 2018

Your auc improves because clusters are created using all the data so when you fit your model on the augmented training set you are implicitly including an information (the clustering labels) correlated with the response variable in the test set.

Prakash Jhawar
Prakash Jhawar 14 May, 2018

Nice Article ! Much appreciation to the author. But, I am wondering we are running cluster in the over all data and then dividing the dataset in training and testing. No the testing data is not completely unseen to the model developed using training data. Because the clusters in the testing data was designed with the help of training data and model knows what to map to cluster 1,2, 3,..... ! Rather, if we can cluster only basis the training data and then predict the cluster in test data and run the model and still if we get better accuracy we can certainly conclude the same the clustering is enhancing the accuracy. Can you clarify the doubt please ?

Machine Learning
Become a full stack data scientist