Comparative analyses of statistical classification methods / by Zachary Barnhart

In today’s computer-driven world, there are many types of data that need to be parsed through. Because manually processing information can be time-consuming and expensive, we need efficient methods to process a large amount of information. One application of processing information is classifying observations into categories based on specific features. With classification methods, we train a computer to accept specific input predictors, and the output is the class the computer thinks those observations belong in. In this thesis, we will examine the performance of several classification methods under various scenarios. We compare the performance of each method on simulated mixtures of distributions by using the misclassification rate. Furthermore, we extend our findings from the simulation studies by performing each classification method on several real-world datasets.