Visual Exploration and Comparison of Multi-Classification Model with High Number of Classes
Sai Praneeth Gudala, Yash Jitendrabhai Bhadiyadra, Aung Phyo, Durga Silva Lokesh Telaprolu
Department of Computer Science
Faculty Supervisor: Shahrukh Humayoun
The recent advancements in machine/deep learning (ML/DL) have motivated researchers and practitioners to create, and to manage classification models dealing with hundreds of classes, e.g., the case of image datasets. The exploration of ML/DL models with hundreds of classes face two main challenges: 1) to identify the root cause of a problem that occurs during the learning phase such as misclassification of instances, and 2) to compare of models for the same dataset with hundreds of classes.
In this work, we aim to address these challenges by applying visual analytics techniques. Our visual analytics tool assists machine learning experts in identifying the root causes of misclassification in models with hundreds of classes. Additionally, it facilitates the comparison of different models through interactive visualizations. By leveraging this visualization approach, ML experts can gain deeper insights into model performance, streamline the debugging process, and make more informed decisions when managing and refining complex classification models.