Statistical Analysis and Clustering of Tinnitus related Data with respect to the perceived Symptoms

Abstract

A lot of people suffer from chronic diseases. Sometimes, cause or treatment methods are not well-known or well-researched. In order to gain a better understanding, personal data is collected. Afterwards, the collected data can be analyzed using a huge set of different techniques. One of these techniques is cluster analysis. Cluster analysis is mostly used to group a data set into multiple smaller sets. The new identified sets contain only data points which are similar to each other. In this thesis, we are evaluating popular clustering algorithms and test them on our tinnitus related data set. Data points were collected beforehand from applications like TrackYourTinnitus and TrackYourHearing. Both applications provided questions with predefined answers. The purpose of these applications will be addressed in this work. Also, all relevant algorithms are explained in detail. Furthermore, advantages and disadvantages of each approach are discussed. One approach that is considered in this work is Latent Class Analysis. It is a special case of Structural Equation Modeling which in general is a statistical modeling technique primarily used in behavioral science. We also make use of two well-known machine learning clustering algorithms. The two other approaches are K-Means/K-Modes clustering and Density-Based Spatial Clustering of Applications with Noise (DBSCAN). K-Means/K-Modes clustering and DBSCAN are both unsupervised machine learning algorithm whose goal it is to find appropriate clusters. In the end of this work, the results are evaluated. The mayor goal of this work is to gain a better understanding of tinnitus perception for individuals and their clusters.