Affiliations: [a] Graduate Schools of Electrical Engineering and Applied Computer Science, Federal Technological University of Parana (UTFPR), Avenida Sete de Setembro, Curitiba, Paraná, Brazil | [b] Post-Graduate Program in Informatics, Pontifical Catholic University of Parana (PUCPR), Rua Imaculada Conceição, Curitiba, Paraná, Brazil
Correspondence:
[*]
Corresponding author. Cecilia H. Vallejos de Schatz and Fabio K. Schneider, Graduate Schools of Electrical Engineering and Applied Computer Science, Federal Technological University of Parana (UTFPR), Avenida Sete de Setembro, 3165, Curitiba, Paraná, Brazil. Tel.: +55 41 3310 4790; Fax: +55 41 3310 4432; E-mails: [email protected] (C.H. Vallejos de Schatz), [email protected] (F.K. Schneider).
Abstract: Prediction of physiological vital signs allow early detection of abnormal conditions in chronic or elderly patients. In this paper, an artificial intelligent tool is proposed using fuzzy logic and recurrent neural networks for definition and forecast of patient’s clinical condition by using five clinical parameters. For this purpose, a fuzzy model and six artificial neural networks were tested and compared. The fuzzy logic-based proposed first phase of the tool permits the analysis of the current state of the patient, which allows the training of the artificial neural network. In the second phase, two Elman networks MISO, two Elman networks MIMO, as well as two Nonlinear Autoregressive with eXogenous inputs (NARX) neural networks are evaluated with and without pruning. After analyzing the six proposed networks, it was verified that the pruned NARX model presents the best performance and for this reason was chosen to build the Fuzzy-NARX (FNARX) prediction tool. Finally, the FNARX solution implemented was tested by utilizing unseen data from thirty patients with very high accuracy. More tests made with higher prediction periods demonstrate a slight decrease in the performance. Simulation results demonstrate the effectiveness of the FNARX model proposed, which has a very good performance in predicting patient conditions, and it is a useful tool for preventive medicine.
