Unsupervised Machine Learning Models for User Segmentation in a Mobile App : Evaluating the Effectiveness
No Thumbnail Available
Files
Date
2023
Authors
Journal Title
Journal ISSN
Volume Title
Company
URL
Shelf Location
Date
Abstract
AI-GENERATED ABSTRACT: Abstract: Context: User segmentation is an effective tool to understand your users' needs and make data-driven decisions. With limited resources many startups have. Would user clustering with unsupervised machine learning models be effective? Objectives: The thesis consists of a literature review, two interviews, and data analysis. It evaluates how effective the different unsupervised models are in application to user segmentation in a mobile application. Methods: a convenience sample interview was conducted with the chief executives of a startup to get an understanding of primary goals and objectives. Data were analyzed with the application of dimension reduction (PCA, t-SNE, UMAP, correlation coefficient, variance threshold), clustering (K-Means, DBSCAN), and supervised models for predictive analysis (Random Forest, Lasso, Logistic Regression). Results: It was possible to identify 4 different clusters of users within the app with unique behavior. Conclusion: The application of PCA, K-Means, and Random Forest was the most effective for a highly dimensional dataset. This user segmentation was valuable, but not new to the chief executives to the company. Meaningful insights were drawn from data analysis. Keywords: user segmentation, unsupervised learning, machine learning, mobile application, PCA, K-Means clustering, Random Forest, data analysis, startups, user behavior clustering