Learning Rate in Supervised Learning Models – The Case of Digit Identification

Abstract

In this paper, I explore the application of machine learning models for handwritten digit identification. The goal is to train and predict multiple machine learning models with a dataset of handwritten digits. I vary the learning rate of the model to investigate the relationship between learning rate and prediction accuracy. I find that the prediction accuracies for both training and test data increase between learning rate of 0 and 0.1, and then fluctuate around a constant beyond learning rate of 0.1. The fluctuations are greater for pairs of digits that are more difficult to distinguish.