Att klassificera med Support Vector Machines - En introduktion från teori till analys

Sammanfattning: The purpose of this paper is to give a short introduction to support vector machines. The paper intends to cover the full process from theory to analysis in the binary classification case. The analysis intends to yield an understanding of how the method can be used in practice, and how models can be fitted in a way that maximizes the desired performance measurement. This paper introduces the theory behind classification in the most basic case with perfect linear separability. Subsequently, methods to expand the model to manage more realistic scenarios of non-linear separability are introduced. Furthermore, the model selection process is detailed, and common performance measurements are proposed. The dataset, Wisconsin Diagnostic Breast Cancer, is used for the analysis. Numerous models are trained to predict tumors as malignant or benign. The final model performs in line with predetermined requirements, correctly classifying as many malignant tumors as possible. The final model attains a sensitivity of 0.9836, which equals one false negative prediction.