“…Among the many possible SI measures that could be used, we adopt Thornton's SI [52], a competitive measure for assessing class separation originally introduced for supervised classification problems [4], [10], [23]. Also called the geometric separability index (GSI), Thornton's measure has seen wide application in health, robotics, geology and other fields [42], [14], [49], [54], [47]. We use the following simple unsupervised modification of the GSI [52], computed on the merged training and test samples for the batch: We performed a simulation of HMM-FLDA that emulates the experimental study presented in Section IV, but with ground truth event states.…”