Feature Selection for SVM-Based Vascular Anomaly Detection

This work explores feature selection to improve the performance in the vascular anomaly detection domain. Starting from a previously defined classification framework based on Support Vector Machines (SVM), we attempt to determine features that improve classification performance and to define guidelines for feature selection. Three different strategies were used in the feature selection stage, while a Density Level Detection-SVM (DLD-SVM) was used to validate the performance of the selected features over testing data. Results show that a careful feature selection results in a good classification performance. DLD-SVM shows a poor performance when using all the features together, owing to the curse of dimensionality.