How to do Agglomerative Clustering in R?

Recipe Objective

Organised logical groups of information is preferred over unorganised data by people. For example, anyone finds it easier to remember information when it is clustered together by taking its common characteristics into account.

Likewise, A machine learning technique that provides a way to find groups/clusters of different observations within a dataset is called Clustering. In this technique due to the absence of response variable, it is considered to be an unsupervised method. This implies that the relationships between 'n' number of observations is found without being trained by a response variable. The few applications of Clustering analysis are ​

1. Customer segmentation: process for dividing customers into groups based on similar characteristics.
2. Stock Market Clustering based on the performance of the stocks
3. Reducing Dimensionality

There are two most common Clustering algorithm that is used: ​

1. KMeans Clustering: commonly used when we have large dataset
2. Heirarchical or Agglomerative Clustering: commonly used when we have small dataset

Agglomerative Clustering is an unsupervised machine learning technique that aims to groups the unlabeled dataset by building a heirarcy of clusters. It is relatively slow compared to heirarchichal clustering. There are two types of Heirarchical clustering algorithm: Divisive (top-down appraoch) and Agglomerative (bottom-up approach). ​

The most commonly used is the agglomerative algorithm. In this algorithm, data is split into n clusters where n is the observations in the dataset initially. The next step is the calculation of euclidean distances between data points. Then, the number of clusters are reduced because the two clusters merges into one in an iterative manner considering the distances between them. This process of merging stops when only one cluster remains. The hierarchy of clusters is represented by Dendrogram (a tree-like structure) ​

This recipe demonstrates Heirarchical Clustering using Agglomerative algorithm on a real-life Mall dataset to carry out customer segmentation in R-language. ​

STEP 1: Importing Necessary Libraries

STEP 2: Loading the Dataset

Dataset description: It is a basic data about the customers going to the supermarket mall. This can be used for customer segmentation. There are 200 observations(customers) and no missing data.

It consists of four columns ie. measured attrutes: ​

1. CustomerID is the customer identification number.
2. Gender is Female and Male.
3. Age is the age of customers.
4. Annual Income (k) is the annual income of clients in thousands of dollars.
5. Spending Score (1-100) is the spending score assigned by the shopping center according to the customer's purchasing behavior

Observations: 200
Variables: 5
$ CustomerID              1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 1...
$ Gender                  Male, Male, Female, Female, Female, Female, ...
$ Age                     19, 21, 20, 23, 31, 22, 35, 23, 64, 30, 67, ...
$ Annual.Income..k..      15, 15, 16, 16, 17, 17, 18, 18, 19, 19, 19, ...
$ Spending.Score..1.100.  39, 81, 6, 77, 40, 76, 6, 94, 3, 72, 14, 99,...

For the simplicity of demonstrating heirarchichal Clustering with visualisation, we will only consider two measured attributes (Age and Annual Income). ​

STEP 3: Data Preprocessing (Scaling)

This is a pre-modelling step. In this step, the data must be scaled or standardised so that different attributes can be comparable. Standardised data has mean zero and standard deviation one. we do thiis using scale() function

Note: Scaling is an important pre-modelling step which has to be mandatory

Age		Annual.Income..k..
-1.4210029	-1.734646
-1.2778288	-1.734646
-1.3494159	-1.696572
-1.1346547	-1.696572
-0.5619583	-1.658498
-1.2062418	-1.658498

STEP 4: Calculating the distances between the observations/datapoints

We use the dist() function to carry this out task. It calculates the euclidean distances between the datapoints. We create an object 'distances' to store this information.

0.143174096235284 0.0810822191043143 0.288868323164455 0.862412934253182 0.227861432257651 1.15107392658714

STEP 5: Performing Heirarchical clustering

We will use hclust() function in cluster library in R to perform this. The two arguements used below are:

1. Distances between the point
2. method of evaluation: Ward's method (Ward.D)

STEP 6: Data Visualization using scatter plot with specified number of clusters

We use cutree() function in cluster library to specify the number of clusters to be formed. This function cuts the dendrogram in such a way that only the specified number of clusters are obtained. In our case, we will use 5 number of clusters.

We use clusplot to plot theclusters in a scatter plot w.r.t Age and Income

This plot helps us to analyse the different clusters of customers formed so that we can target the respective clusters seperately in our marketing strategy.