Josie Wallin

Document Type

Thesis - University Access Only

Award Date

2010

Degree Name

Master of Science (MS)

Department / School

Mathematics and Statistics

Abstract

A business must know who its customers are in order to meet its goals and operate efficiently. The products and services available from a business, along with marketing strategy, should reflect a business' current or goal customers in order to maximize profits and efficiency. This problem can be solved with a data mining technique called cluster analysis. Cluster analysis is a knowledge discovery tool capable of summarizing hundreds of thousands of observations on several variables by finding groups within data. The main objective of this thesis was to determine how many different types of customers applied for student loans at a particular student loan company. The original dataset consisted of more than 470,000 student-loan applications received by the company between April 2008 and March 2009. Different clustering methods were employed in order to answer these questions, including hierarchical agglomerative and k-means clustering. The range of possible cluster solutions was found to be two to six, and each of these solutions was run using k-means clustering on the entire original dataset. After comparing the results, the most appropriate solution was found to be three clusters. The three cluster solution was validated using several different methods which included running the three cluster solution using k-means clustering on random samples of the original data and also on one year prior's data (April '07 to March '08 student applicants) and the most recent year's data (April '09 to March '10). After each method, the results were analyzed to determine if three clusters similar to the original solution were found. The researcher found that the 3 clusters, representing three typical types of student loan applicants, were consistently detected and validated in different years' data as well as different subsets of the original data. The analysis also showed some evidence of a new type of customer emerging from the most recent year's data which might possibly be explained by the Great Recession of 2008 and 2009. In general, it was found that cluster analysis can successfully be used to detect customer segments of student loan applicants, which could be potentially useful for targeted marketing or corporate decision making.

Library of Congress Subject Headings

Student loans

Cluster analysis

Market segmentation

Format

application/pdf

Number of Pages

55

Publisher

South Dakota State University

Download

Share

COinS