In the previous chapter, we saw how to represent data in a tabular format, create features and target matrices, preprocess data, and choose the algorithm that best suits the problem at hand. We also saw how the scikit-learn API works and why it is easy to use.

The main objective of this chapter is to solve a real-world case study, where the students 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 such a problem. At the end of this chapter, the students 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 machine-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 find meaning in the structure of large data in order to draw decisions.

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

Moreover, 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 greatly differ from the data points in other clusters.

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.

Datasets in repositories may contain raw, partially preprocessed, or preprocessed data. To use any of these datasets, ensure that you read the specifications of the data available to understand the process that needs to be followed to model the data effectively.

Once data has been revised generically to ensure that it can be used for the desired purpose, it is time to load the dataset and use data visualization to further understand it. Data visualization is not a requirement for developing a machine-learning project, especially when dealing with datasets with hundreds or thousands of features. However, it has become an integral part of machine learning, mainly for visualizing the following:

Its popularity in the tasks detailed previously is explained by the fact that the human brain processes information easily when it is presented as charts or graphs, which allows us to have a general...

The k-means algorithm is used for data without a labeled class. It involves dividing the data into K number of subgroups. The classification of data points into each group is done based on similarity, as explained before, which for this algorithm is measured by the distance from the center (centroid) of the cluster. The final output of the algorithm are the data points related to a cluster and the centroid of each cluster, which can be used to label new data in the same clusters.

The centroid of each cluster represents a collection of features that can be used to define the nature of the data points that belong there.

Understanding the Algorithm

The k-means algorithm works through an iterative process that involves the following steps:

Figure 2.6: A formula minimizing the Euclidean distance

Steps 2 and 3 are repeated in an iterative process, until a criterion is met. The criterion can be as follows:

• The number of iterations defined.

• The data points do not change from cluster...

The mean-shift algorithm works by assigning each data point a cluster based on the density of data points in the data space, also known as the mode in a distribution function. Contrary to the k-means algorithm, the mean-shift algorithm does not require you to specify the number of clusters as a parameter.

The algorithm works by modeling the data points as a distribution function, where high-density areas (high concentration of data points) represent high peaks. Then, the general idea is to shift each data point until it reaches its nearest peak, which becomes a cluster.

Understanding the Algorithm

The first step of the mean-shift algorithm is the representation of the data points as a density distribution. To do so, the algorithm builds upon the idea of Kernel Density Estimation (KDE), which is a method used to estimate the distribution of a set of data:

Figure 2.10: An image depicting the idea behind Kernel Density Estimation

In the preceding diagram, the red dots represent...

The density-based spatial clustering of applications with noise (DBSCAN) algorithm groups together points that are close to each other (with many neighbors) and marks those points that are further away with no close neighbors as outliers.

According to this, and as its name states, the algorithm classifies data points based on the density of all data points in the data space.

Understanding the Algorithm

The DBSCAN algorithm requires two main parameters: epsilon and the minimum number of observations.

Epsilon, also known as eps, is the maximum distance that defines the radius within which the algorithm searches for neighbors. The minimum number of observations, on the other hand, refers to the number of data points required to form a high density area (min_samples). However, the latter is optional in scikit-learn as the default value is set to 5:

Figure 2.13: An illustration of how the DBSCAN algorithm classifies data into clusters

In the preceding diagram, the blue dots are assigned...

After applying a clustering algorithm, it is necessary to evaluate how well the algorithm has performed. This is especially important when it is difficult to visually evaluate the clusters, for example, when there are several features.

Usually, with supervised algorithms, it is easy to evaluate the performance by simply comparing the prediction of each instance with its true value (class). On the other hand, when dealing with unsupervised models, it is necessary to pursue other strategies. In the specific case of clustering algorithms, it is possible to evaluate performance by measuring the similarity of the data points that belong to the same cluster.

Data problems where the input data is unrelated to a labeled output is handled using unsupervised learning. The main objective of such data problems is to understand the data by finding patterns that, in some cases, can be generalized to new instances. In this context, this chapter covered clustering algorithms, which work by aggregating similar data points into clusters, while separating data points that greatly differ. After this, the chapter covered data visualization tools that can be used to analyze problematic features during data preprocessing. We also saw how to apply different algorithms to the dataset and compare their performance to choose the one that best fits the data. Two different metrics for performance evaluation, the Silhouette Coefficient metric and the Calinski-Harabasz index, were also discussed in light of the inability to represent all of the features in a plot, and thereby graphically evaluate performance on scikit-learn. However, it is important to understand...