In Chapter 5, we introduced clustering, an unsupervised learning approach to identify the underlying structure in data and grouping points based on similarity. These groups (known as clusters) should be homogeneous and distinct. In other words, the members within a group should be very similar to each other and very distinct from members of any other group.
From an applied perspective, the ability to segment members into groups based on similarity and without any guidance from labels is very powerful. For example, such a technique could be applied to find different consumer groups for online retailers, customizing a marketing strategy for each of the distinct groups (i.e., budget shoppers, fashionistas, sneakerheads, techies, audiophiles, etc.). Group segmentation could improve targeting in online advertising and improve recommendations in recommender systems for movies, music, news, social networking, dating, etc.
In this chapter, we will build an applied unsupervised learning solution using the clustering algorithms from the previous chapter—more specifically, we will perform group segmentation.
For this chapter, we will use loan data from Lending Club, a US peer-to-peer lending company. Borrowers on the platform can borrow between $1,000 to $40,000 in the form of unsecured personal loans, for a term of either three or five years.
Investors can browse the loan applications and choose to finance the loans based on the credit history ...