Search

The purpose of this work is to test the efficiency of specific intelligent classification algorithms when dealing with the domain of stroke medical diagnosis. The dataset consists of patient records of the ”Acute Stroke Unit”, Alexandra Hospital, Athens, Greece, describing patients suffering one of 5 different stroke types diagnosed by 127 diagnostic attributes / symptoms collected during the first hours of the emergency stroke situation as well as during the hospitalization and recovery phase of the patients. Prior to the application of the intelligent classifier the dimensionality of the dataset is further reduced using a variety of classic and state of the art dimensionality reductions techniques so as to capture the intrinsic dimensionality of the data. The results obtained indicate that the proposed methodology achieves prediction accuracy levels that are comparable to those obtained by intelligent classifiers trained on the original feature space.

The knowledge of the software quality can allow an organization to allocate the needed resources for the code maintenance. Maintaining the software is considered as a high cost factor for most organizations. Consequently, there is need to assess software modules in respect of defects that will arise. Addressing the prediction of software defects by means of computational intelligence has only recently become evident. In this paper, we investigate the capability of the genetic programming approach for producing solution composed of decision rules. We applied the model into four software engineering databases of NASA. The overall performance of this system denotes its competitiveness as compared with past methodologies, and is shown capable of producing simple, highly accurate, tangible rules.