Interactive Visualization Support for Comparing and Analysis of Multi-class Classifier Problems in Machine Learning
By: Nina Mir
Department: Computer Science
Faculty Advisor: Dr. Shahrukh Humayoun
Classification problems have important applications in different disciplines. Our project provides a tool that could be used by machine learning analysts to compare and contrast the results of multiple multi-class ML models, where the number of classes is approaching 1000.
Displaying such complex abstract information could cause problems like visual clutter that makes it difficult to gain insight. It is also of interest for ML analysts to be able to compare the results of numerous models simultaneously. This issue, combined with the inherent problem of classification problems involving several hundreds of classes, can easily become prohibitive for any data visualization system. We have developed a web app that offers a radial view solution to this specific problem. The results of our work, including a working prototype, are offered in this work.