Analyzing Decision Trees to Understand MNIST Misclassification
Alexis Comeau, Christopher McDonald · 2019
The MNIST dataset of handwritten digits is a classic tool for the study of machine learning. In this paper we use MNIST and a decision tree classifier to make a case study of misclassification for the digits three and five, analyzing the nodes of the decision tree plot to understand the process of classification. The paper concludes with a discussion of possibilities for our research and a look at some related, higher-level research that has been done.