Cluster analysis is an analytical method for classifying data with many different items into similar items.
In marketing, it is used for market research and analysis of customer information.
In this article, we will look at types of cluster analysis, application examples, and points to note when using cluster analysis.
What is cluster analysis?
Cluster analysis is an analytical method for classifying data with many different items into similar items.
Cluster analysis is often used in marketing for market research and customer information analysis.
Types of cluster analysis can be broadly divided into “hierarchical” and “non-hierarchical”.
Each has advantages and disadvantages, so it is necessary to use them appropriately depending on the type of data and the purpose of analysis.
Cluster analysis is a powerful method for understanding trends and characteristics in large amounts of data.
However, it is not a universal analysis method, so it is important to understand its limitations when using it.
Types of cluster analysis
For cluster analysis,
・Hierarchical cluster analysis
・Non-hierarchical cluster analysis
There are two types.
Let’s take a look at what each of these are.
●Hierarchical cluster analysis
Hierarchical cluster analysis starts with the most similar items and gradually moves to the least similar items.
The image of hierarchical cluster analysis is as shown in the image below.
[Image diagram of hierarchical cluster analysis]
In the above example, A and B, which have the highest degree of similarity, are first classified into the same cluster, and then C and D are classified into the same cluster.
Then, the AB cluster and the CD cluster are classified as one large cluster, and the distant E is finally classified there when the whole cluster becomes one.
In hierarchical cluster analysis, the processes that are classified into clusters are shown as a “dendrogram” on the right side of the figure above, similar to a tournament diagram.
For this reason, the advantage is that you can decide how many clusters to divide into after analysis while looking at the dendrogram.
In the example above, you could have three clusters: “AB”, “CD”, and “E”, or you could have two clusters: “ABCD” and “E”.
However, hierarchical cluster analysis also has major drawbacks.
The problem is that the similarity between all elements is calculated using brute force, which increases the amount of calculation, and the calculation may become difficult when the amount of data increases.
Therefore, when analyzing a large amount of data such as big data, it is common to use non-hierarchical cluster analysis, as described below.
●Non-hierarchical cluster analysis
Non-hierarchical cluster analysis does not use brute-force calculations like hierarchical cluster analysis, but simplifies calculations.
For this reason, this is mainly used when analyzing big data.
However, in non-hierarchical cluster analysis, it is necessary to decide how many clusters to classify into before calculation.
Therefore, trial and error may be required to obtain the appropriate results, such as performing the calculation several times while changing the number of clusters.
Also, when calculating, initial values must be determined.
The results may vary depending on this initial value.
For this reason, it may be necessary to perform trial and error calculations by changing the initial values several times.
Application examples of cluster analysis
Cluster analysis is
・Market research
・Customer survey
・Customer information such as credit cards and point cards
A popular way to classify them is as follows.
Market research, customer surveys, and card customer information data contain a variety of information such as age, gender, and address.
For example, suppose that by performing cluster analysis on these data, we obtain the following clusters.
・Luxury product oriented type
・Fashion-seeking type
・Low interest type
・Maintenance type
This classification makes it easier to understand trends in the overall data.
Additionally, in the case of customer surveys and card customer information, it is also possible to send DMs introducing products that match the customer’s interests for each cluster.
Things to keep in mind when using cluster analysis
Things to keep in mind when using cluster analysis:
・Cluster analysis cannot be called “objective”
・Needs to be used in combination with other analysis methods
I can give you two.
●Cluster analysis cannot be called “objective”
This is especially true for non-hierarchical cluster analysis, which is often used in practice.
In nonhierarchical cluster analysis, the number of clusters must be determined in advance.
Also, the results may vary depending on the initial values.
Therefore, non-hierarchical cluster analysis cannot be called an “objective analysis”.
The analysis results will include the judgment of the analyst.
It is important to be careful not to treat cluster analysis as objective.
●Needs to be used in conjunction with other analysis methods
In addition, cluster analysis only performs “classification”.
Cluster analysis cannot reveal what kind of regularity or causal relationship caused the classification.
Therefore, when considering actual actions such as business improvements, it is important not to make decisions based solely on the results of cluster analysis.
It is important to make decisions after obtaining more multifaceted analysis results by using other analysis methods such as regression analysis and correlation analysis.
summary
◆ Cluster analysis is an analytical method for classifying many different things into similar ones ◆ There are two types of cluster analysis: “hierarchical” and “non-hierarchical” ◆ Cluster analysis is used to analyze market research and customer It is used to classify questionnaires, card customer information, etc. ◆ It is important to understand that cluster analysis cannot be said to be objective, and that it only performs classification.