Choosing Appropriate Performance Measure for Binary Clas-sification Problem
Name
Egert Georg Teesaar
Abstract
Machine learning is a big part of artificial intelligence which tries to imitatethe learning process what every living organism uses in everyday life to deal witharisen problems. It uses available data about the problem of interest to learn anddetect patterns so it can build a model which could be helpful in overcoming similarproblems in the future.One branch of machine learning is binary classification. It specializes in problemswhere there are only two possible outcomes also know as classes. Therefore themodel which has been trained to solve such problems can only predict class Aor class B. This in turn raises question how can one know if the given modelis appropriate to deal with such problems. One way to evaluate this, is to useperformance meaasures.This thesis focuses on how to choose appropriate performance measure forbinary classification problems. It brings out different measures and providesquestions which try to discover the purpose of the model and the context in whichit was trained. There have been published many works which give an insight intoperformance measures and their characteristics but they require the reader toalready be familiar with the topic and therefore leave out risks and shortcomingsof certain measures. They also don’t provide a concrete manual on how to choosea performance measure.This work tries to help people, who lack deeper knowledge about machinelearning, to discover appropriate measure for the problem in hand.
Graduation Thesis language
Estonian
Graduation Thesis type
Bachelor - Computer Science
Supervisor(s)
Mari-Liis Allikivi, Meelis Kull
Defence year
2018