In the previous chapter, we learned how to represent data in a tabular format, created features and target matrices, pre-processed data, and learned how to choose the algorithm that best suits the problem at hand. We also learned how the scikit-learn API works and why it is easy to use, as well as the difference between supervised and unsupervised learning.

This chapter focuses on the most important task in the field of unsupervised learning: clustering. Consider a situation in which you are a store owner wanting to make a targeted social media campaign to promote selected products to certain customers. Using clustering algorithms, you would be able to create subgroups of your customers, allowing you to profile those subgroups and target them accordingly. The main objective of this chapter is to solve a case study, where you will implement three different unsupervised learning solutions. These different applications serve to demonstrate the uniformity of the scikit-learn API, as well as to explain the steps taken to solve machine learning problems. By the end of this chapter, you will be able to understand the use of unsupervised learning to comprehend data in order to make informed decisions.

Clustering is a type of unsupervised learning technique where the objective is to arrive at conclusions based on the patterns found within unlabeled input data. This technique is mainly used to segregate large data into subgroups in order to make informed decisions.

For instance, from a large list of restaurants in a city, it would be useful to segregate the data into subgroups (clusters) based on the type of food, quantity of clients, and style of experience, in order to be able to offer each cluster a service that's been configured to its specific needs.

Clustering algorithms divide the data points into n number of clusters so that the data points in the same cluster have similar features, whereas they differ significantly from the data points in other clusters.

Clustering Types

Clustering algorithms can classify data points using a methodology that is either hard or soft. The former designates data points completely to a cluster, whereas the latter method calculates the probability of each data point belonging to each cluster. For example, for a dataset containing customer's past orders that are divided into eight subgroups (clusters), hard clustering occurs when each customer is placed inside one of the eight clusters. On the other hand, soft clustering assigns each customer a probability of belonging to each of the eight clusters.

Considering that clusters are created based on the similarity between data points, clustering algorithms can be further divided into several groups, depending on the set of rules used to measure similarity. Four of the most commonly known sets of rules are explained as follows:

• Connectivity-based models: This model's approach to similarity is based on proximity in a data space. The creation of clusters can be done by assigning all data points to a single cluster and then partitioning the data into smaller clusters as the distance between data points increases. Likewise, the algorithm can also start by assigning each data point an individual cluster, and then aggregating data points that are close by. An example of a connectivity-based model is hierarchical clustering.
• Density-based models: As the name suggests, these models define clusters by their density in the data space. This means that areas with a high density of data points will become clusters, which are typically separated from one another by low-density areas. An example of this is the DBSCAN algorithm, which will be covered later in this chapter.
• Distribution-based models: Models that fall into this category are based on the probability that all the data points from a cluster follow the same distribution, such as a Gaussian distribution. An example of such a model is the Gaussian Mixture algorithm, which assumes that all data points come from a mixture of a finite number of Gaussian distributions.
• Centroid-based models: These models are based on algorithms that define a centroid for each cluster, which is updated constantly by an iterative process. The data points are assigned to the cluster where their proximity to the centroid is minimized. An example of such a model is the k-means algorithm, which will be discussed later in this chapter.

In conclusion, data points are assigned to clusters based on their similarity to each other and their difference from data points in other clusters. This classification into clusters can be either absolute or variably distributed by determining the probability of each data point belonging to each cluster.

Moreover, there is no fixed set of rules to determine similarity between data points, which is why different clustering algorithms use different rules. Some of the most commonly known sets of rules are connectivity-based, density-based, distribution-based, and centroid-based.

Applications of Clustering

As with all machine learning algorithms, clustering has many applications in different fields, some of which are as follows:

• Search engine results: Clustering can be used to generate search engine results containing keywords that are approximate to the keywords searched by the user and ordered as per the search result with greater similarity. Consider Google as an example; it uses clustering not only for retrieving results but also for suggesting new possible searches.
• Recommendation programs: It can also be used in recommendation programs that cluster together, for instance, people that fall into a similar profile, and then make recommendations based on the products that each member of the cluster has bought. Consider Amazon, for example, which recommends more items based on your purchase history and the purchases of similar users.
• Image recognition: This is where clusters are used to group images that are considered to be similar. For instance, Facebook uses clustering to help suggest who is present in a picture.
• Market segmentation: Clustering can also be used for market segmentation to divide a list of prospects or clients into subgroups in order to provide a customized experience or product. For example, Adobe uses clustering analysis to segment customers in order to target them differently by recognizing those who are more willing to spend money.

The preceding examples demonstrate that clustering algorithms can be used to solve different data problems in different industries, with the primary purpose of understanding large amounts of historical data that, in some cases, can be used to classify new instances.

As part of the process of learning the behavior and applications of clustering algorithms, the following sections of this chapter will focus on solving a real-life data problem using the Wholesale Customers dataset, which is available at the UC Irvine Machine Learning Repository.

Note

Datasets in repositories may contain raw, partially pre-processed, or pre-processed data. To use any of these datasets, ensure that you read the specifications of the data that's available to understand the process that needs to be followed to model the data effectively, or whether it is the right dataset for your purpose.

For instance, the current dataset is an extract from a larger dataset, as per the following citation:

The dataset originates from a larger database referred on: Abreu, N. (2011). Analise do perfil do cliente Recheio e desenvolvimento de um sistema promocional. Mestrado em Marketing, ISCTE-IUL, Lisbon.

In the following section, we will analyze the contents of the dataset, which will then be used in Activity 2.01, Using Data Visualization to Aid the Pre-processing Process. To download a dataset from the UC Irvine Machine Learning Repository, perform the following steps:

1. Access the following link: http://archive.ics.uci.edu/ml/datasets/Wholesale+customers.
2. Below the dataset's title, find the download section and click on Data Folder.
3. Click on the Wholesale Customers data.csv file to trigger the download and save the file in the same path as that of your current Jupyter Notebook.

   Note

   You can also access it by going to this book's GitHub repository: https://packt.live/3c3hfKp

Understanding the Dataset

Each step will be explained generically and will then be followed by an explanation of its application in the current case study (the Wholesale Customers dataset):

1. First of all, it is crucial to understand the way in which data is presented by the person who's responsible for gathering and maintaining it.

   Considering that the dataset of the case study was obtained from an online repository, the format in which it is presented must be understood. The Wholesale Customers dataset consists of a snippet of historical data of clients from a wholesale distributor. It contains a total of 440 instances (each row) and eight features (each column).

2. Next, it is important to determine the purpose of the study, which is dependent on the data that's available. Even though this might seem like a redundant statement, many data problems become problematic because the researcher does not have a clear view of the purpose of the study, and hence the pre-processing methodology, the model, and the performance metrics are chosen incorrectly.

   The purpose of using clustering algorithms on the Wholesale Customers dataset is to understand the behavior of each customer. This will allow you to group customers with similar behaviors into one cluster. The behavior of a customer will be defined by how much they spent on each category of product, as well as the channel and the region where they bought products.

3. Subsequently explore all the features that are available. This is mainly done for two reasons: first, to rule out features that are considered to be of low relevance based on the purpose of the study or that are considered to be redundant, and second, to understand the way the values are presented to determine some of the pre-processing techniques that may be needed.

   The current case study has eight features, each one of which is considered to be relevant to the purpose of the study. Each feature is explained in the following table:

   

Figure 2.1: A table explaining the features in the case study

In the preceding table, no features are to be dismissed, and nominal (categorical) features have already been handled by the author of the dataset.

As a summary, the first thing to do when choosing a dataset or being handed one is to understand the characteristics that are visible at first glance, which involves recognizing the information available, then determining the purpose of the project, and finally revising the features to select those that will be part of the study. After this, the data can be visualized so that it can be understood before it's pre-processed.