Fundamentals and Methods of Machine and Deep Learning

3Model Evaluation

Ravi Shekhar Tiwari

University of New South Wales, Kensington, Sydney, Australia

Abstract

We are in a technology-driven era, where 70% of our activities are directly dependent on technology and the remaining 30% is indirectly dependent. During the recent time, there was a breakthrough in computing power and in the storage devices—which gave rise to technology such as cloud, Gpu, and Tpu—as a result of our storage and processing capacity increased exponentially. The exponential increase in storage and computing capabilities enables the researcher to deploy Artificial Intelligence (A.I.) in a real-world application.

A.I. in real-world application refers to the deployment of the trained Machine Learning (ML) and Deep Learning (DL) models, to minimize the human intervention in the process and make the machine self-reliant. As we all know, for every action, there is a positive and negative reaction. These breakthroughs in the technology lead to the creation of a variety of ML/DL models but the researchers were stupefied between the selection of models. They were bewildered which model they should select as to correctly mimic the human mind—the goal of A.I. As most of us are solely dependent on the accuracy metric to justify the performance of our model, but in some cases, the accuracy of the simple model and complex model are almost equivalent. To solve this perplexity, researchers came up with a variety of the metrics which are dependent on the dataset on which ...