Abstract

Abstract

Abstract
The analysis of epileptic seizures is typically performed by visual inspection, limited by interrater variation. Our aim was to differentiate seizures characterized by automatisms with an objective, quantitative movement analysis. In part 1 of this study we found parameters (extent and speed of movement of the wrist and trunk) separating seizures with predominant proximal (hyperkinetic, n=10) and distal (automotor, n=10) limb automatisms (P<0.002). For each movement parameter we used the lowest value recorded for a hyperkinetic seizure in part 1 as the cutoff parameter in part 2 on a consecutive sample of 100 motor seizures. As in part 1, the difference between hyperkinetic and non-hyperkinetic seizures was highly significant (<0.001). When all movement parameters were above the threshold, a hyperkinetic seizure was identified with a probability of 80.8%, but the probability for a non-hyperkinetic seizure to have all parameters above the threshold was only 0.02%.

Abstract
The area of object/people location consists on the identification of their instantaneous movement in a given area of space and during a given time frame. This location takes an important role because it can help rectify trajectories (in the case of autonomous objects) or improve the performance (in the case of sport practices). In this research project, a system based on RFID that automatically rectifies a location and calculates a wheelchair trajectory in a pre-defined route will be presented. For that, a commercial location system was used as well as a wheelchair and two routes: a linear one and a complex route. In order to correct the trajectory a low pass filter was also used. The obtained results showed that this approach decreased 38% of the mean squared error and in more than 7 centimeters the maximum error. In the future this research work will be enclosed in a new project where a multi agent system capable of executing a management of an autonomous wheelchair set in a hospital environment will be developed. © 2011 AISTI.

Abstract
Feature selection is a topic of growing interest mainly due to the increasing amount of information, being an essential task in many machine learning problems with high dimensional data. The selection of a subset of relevant features help to reduce the complexity of the problem and the building of robust learning models. This work presents an adaptation of a recent quadratic programming feature selection technique that identifies in one-fold the redundancy and relevance on data. Our approach introduces a non-probabilistic measure to capture the relevance based on Minimum Spanning Trees. Three different real datasets were used to assess the performance of the adaptation. The results are encouraging and reflect the utility of feature selection algorithms.

Abstract