Statistics of support vector machines with applications

Abstract

The main goal of this dissertation is to attempt to identify whether or not medical breast images contain cancer. This is carried out using the Support Vector Machine, which is a supervised learning algorithm that fits a maximal margin hyperplane on a set of labelled training data. This algorithm is also carried out in conjunction with MayKay’s Evidence Frameworks, which uses Bayesian techniques in order to optimize any unknown parameters which are required for the learning algorithm. We also delve into the origins of this learning algorithm, its foundation and any key results in Stastical Learning Theory. The data used was obtained in part from the Digital Database for Screening Mammography and from the UCI Machine Learning Repository - where a select number of Haralick attributes were extracted from the dataset. A 5-Fold Cross Validation technique was also carried out during the research, in order to validate the results obtained.