Volume 17, Issue 6 (3-2010)                   DMed 2010, 17(6): 29-38 | Back to browse issues page

XML Persian Abstract Print

Download citation:
BibTeX | RIS | EndNote | Medlars | ProCite | Reference Manager | RefWorks
Send citation to:

mehrabi Y, sedehi M, kazemnejad A, joharimajd V, hadaegh F. Artificial Neural Network for joint prediction of Metabolic Syndrome and HOMA-IR: Tehran Lipid and Glucose Study (TLGS) . DMed. 2010; 17 (6) :29-38
URL: http://daneshvarmed.shahed.ac.ir/article-1-7-en.html
, ymehrabi@gmail.com
Abstract:   (19783 Views)

  Background & Objective: Mixed outcomes arise when, in a multivariate model, response variables measured on different scales such as binary and continuous. In a bivariate modeling, when there are mixed response variables, the common methods in classic statistics have shortcomings. This study aimed at designing an appropriate ANN model for modeling and predicting the bivariate mixed responses including metabolic syndrome and HOMA-IR.

  Materials & Methods: A total of 347 participants from the Cohort section of the Tehran Lipid and Glucose Study (TLGS) were studied. The subjects were free of metabolic syndrome, according to the ATPIII criteria, at the beginning. Demographic characteristics, history of coronary artery disease, body mass index, waist circumference, LDL, HDL, total cholesterol, triglyceride, fasting and 2 hours blood sugar, smoking history, systolic and diastolic blood pressure were measured at the baseline. HOMA-IR and incidence of metabolic syndrome, approximately 3 years after the follow-up, were selected as mixed response variables in designing ANN models. Different ANN models were fitted in to data in two stages. Predictive accuracy was applied to compare the ability of models in prediction. MATLAB software was used for analysis of the data.

  Results: In the first stage, the artificial neural network model with 10 nodes in middle layer, resulted in 65.67 and 69 percent predictive accuracy for test and validation dataset, respectively. In the second stage, the predictive accuracy of SCG, OSS and RP algorithms were 78, 76 and 76 percent, respectively, in validation data set. For test dataset, predictive accuracy of the above algorithms was 78.37, 74.32 and 75.67 percent. These three algorithms showed highest predictive accuracy among 11 different algorithms employed in the study.

  Conclusion: The results of this study indicate that the algorithms SCG, OSS and RP in neural network are the best choices, among those used, for simultaneous prediction of metabolic syndrome and HOMA-IR.

Full-Text [PDF 304 kb]   (2591 Downloads)    
Type of Study: Research |

Send email to the article author

© 2020 All Rights Reserved | Daneshvar

Designed & Developed by : Yektaweb