Software updating is not easy. Operating systems usually require some sort of network connection in order to get its updates, which is fairly easy nowadays. However, in embedded systems, such as an MP3 player or a TV, software updating might be impossible, since they are usually isolated systems. If these kind of products, which currently rely increasingly on software, are more sensitive to bugs, and fixing them after the product is been release is almost impossible, we are forced to find new solutions to keep the availability of this kind of systems. Recent researches have uncovered a way to detect system errors (bugs) during execution, by imbuing error detectors in all of the system¿s variables and train those detectors to accept ¿correct¿ values. However, the fact that this approach applies the mentioned error detectors to all the system¿s variables puts a quite significant overhead in the system. With this research we have proposed to find if it is possible to increase the performance of the existing error detection techniques, by diminish the amount of variables that need to be screened without losing diagnosis quality.