Data Analysis From Scratch With Python: Step By Step Guide

Goals & Uses of Clustering
This is a form of Unsupervised Learning where there are no labels or in many
cases there are no truly correct answers. That’s because there were no correct
answers in the first place. We just have a dataset and our goal is to see the
groupings that have organically formed.
We’re not trying to predict an outcome here. The goal is to look for structures in
the data. In other words, we’re “dividing” the dataset into groups wherein
members have some similarities or proximities. For example, each ecommerce
customer might belong to a particular group (e.g. given their income and
spending level). If we have gathered enough data points, it’s likely there are
aggregates.
At first the data points will seem scattered (no pattern at all). But once we apply
a Clustering algorithm, the data will somehow make sense because we’ll be able
to easily visualize the groups or clusters. Aside from discovering the natural
groupings, Clustering algorithms may also reveal outliers for Anomaly Detection
(we’ll also discuss this later).
Clustering is being applied regularly in the fields of marketing, biology,
earthquake studies, manufacturing, sensor outputs, product categorization, and

---

other scientific and business areas. However, there are no rules set in stone when
it comes to determining the number of clusters and which data point should
belong to a certain cluster. It’s up to our objective (or if the results are useful
enough). This is also where our expertise in a particular domain comes in.
As with other data analysis and machine learning algorithms and tools, it’s still
about our domain knowledge. This way we can look at and analyze the data in
the proper context. Even with the most advanced tools and techniques, the
context and objective are still crucial in making sense of data.