Artificial Neural Network to Modeling Zero-inflated Count Data: Application to Predicting Number of Return to Blood Donation.

Haghani, Sh. and Sedehi, M. and Kheiri, S. (2017) Artificial Neural Network to Modeling Zero-inflated Count Data: Application to Predicting Number of Return to Blood Donation. Journal of research in health sciences, 17 (3). E1-4. ISSN 2228-7809

[img]
Preview
Text
5.pdf

Download (560kB) | Preview

Abstract

BACKGROUND Traditional statistical models often are based on certain presuppositions and limitations that may not presence in actual data and lead to turbulence in estimation or prediction. In these situations, artificial neural networks (ANNs) could be suitable alternative rather than classical statistical methods. STUDY DESIGN  A prospective cohort study. METHODS The study was conducted in Shahrekord Blood Transfusion Center, Shahrekord, central Iran, on blood donors from 2008-2009. The accuracy of the proposed model to prediction of number of return to blood donations was compared with classical statistical models. A number of 864 donors who had a first-time successful donation were followed for five years. Number of return for blood donation was considered as response variable. Poisson regression (PR), negative binomial regression (NBR), zero-inflated Poisson regression (ZIPR) and zero-inflated negative binomial regression (ZINBR) as well as ANN model were fitted to data. MSE criterion was used to compare models. To fitting the models, STATISTICA 10 and, R 3.2.2 was used RESULTS: The MSE of PR, NBR, ZIPR, ZINBR and ANN models was obtained 2.71, 1.01, 1.54, 0.094 and 0.056 for the training and 4.05, 9.89, 3.99, 2.53 and 0.27 for the test data, respectively. CONCLUSIONS The ANN model had the least MSE in both training, and test data set and has a better performance than classic models. ANN could be a suitable alternative for modeling such data because of fewer restrictions.

Item Type: Article
Uncontrolled Keywords: Blood donation Count regression Models Zero-inflated count data Artificial neural network
Subjects: WH Hemic and Lymphatic System
Divisions: Faculty of Health > Department of Epidemiology
Depositing User: Users 1 not found.
Date Deposited: 20 Sep 2017 03:14
Last Modified: 20 Sep 2017 03:14
URI: http://eprints.skums.ac.ir/id/eprint/5359

Actions (login required)

View Item View Item