Measuring confidence of missing data estimation for HIV classification
Computational intelligence methods have been applied to classify pregnant women’s HIV status using demographic data from the South African Antenatal Seroprevalence database obtained from the South African Department of Health. Classification accuracies using a multitude of computational intellige...
Main Author: | |
---|---|
Format: | Others |
Language: | en |
Published: |
2009
|
Online Access: | http://hdl.handle.net/10539/7124 |
id |
ndltd-netd.ac.za-oai-union.ndltd.org-wits-oai-wiredspace.wits.ac.za-10539-7124 |
---|---|
record_format |
oai_dc |
spelling |
ndltd-netd.ac.za-oai-union.ndltd.org-wits-oai-wiredspace.wits.ac.za-10539-71242019-05-11T03:41:20Z Measuring confidence of missing data estimation for HIV classification Mistry, Jaisheel Computational intelligence methods have been applied to classify pregnant women’s HIV status using demographic data from the South African Antenatal Seroprevalence database obtained from the South African Department of Health. Classification accuracies using a multitude of computational intelligence techniques ranged between 60% and 70%. The purpose of this research is to determine the certainty of predicting the HIV status of a patient. Ensemble neural networks were used for the investigation to obtain a set of possible solutions. The predictive certainty of each patients predicted HIV status was computed by giving the percentage of most dominant outputs from the set of possible solutions. Ensembles of neural networks were obtained using boosting, bagging and the Bayesian approach. It was found that the ensemble trained using the Bayesian approach is most suitable for the proposed predictive certainty measure. Furthermore, a sensitivity analysis was done to investigate how each of the demographic variables influenced the certainty of predicting the HIV status of a patient 2009-07-27T12:16:58Z 2009-07-27T12:16:58Z 2009-07-27T12:16:58Z Thesis http://hdl.handle.net/10539/7124 en application/pdf |
collection |
NDLTD |
language |
en |
format |
Others
|
sources |
NDLTD |
description |
Computational intelligence methods have been applied to classify pregnant women’s HIV status
using demographic data from the South African Antenatal Seroprevalence database obtained
from the South African Department of Health. Classification accuracies using a multitude of
computational intelligence techniques ranged between 60% and 70%. The purpose of this
research is to determine the certainty of predicting the HIV status of a patient. Ensemble
neural networks were used for the investigation to obtain a set of possible solutions. The
predictive certainty of each patients predicted HIV status was computed by giving the
percentage of most dominant outputs from the set of possible solutions. Ensembles of neural
networks were obtained using boosting, bagging and the Bayesian approach. It was found that
the ensemble trained using the Bayesian approach is most suitable for the proposed predictive
certainty measure. Furthermore, a sensitivity analysis was done to investigate how each of the
demographic variables influenced the certainty of predicting the HIV status of a patient |
author |
Mistry, Jaisheel |
spellingShingle |
Mistry, Jaisheel Measuring confidence of missing data estimation for HIV classification |
author_facet |
Mistry, Jaisheel |
author_sort |
Mistry, Jaisheel |
title |
Measuring confidence of missing data estimation for HIV classification |
title_short |
Measuring confidence of missing data estimation for HIV classification |
title_full |
Measuring confidence of missing data estimation for HIV classification |
title_fullStr |
Measuring confidence of missing data estimation for HIV classification |
title_full_unstemmed |
Measuring confidence of missing data estimation for HIV classification |
title_sort |
measuring confidence of missing data estimation for hiv classification |
publishDate |
2009 |
url |
http://hdl.handle.net/10539/7124 |
work_keys_str_mv |
AT mistryjaisheel measuringconfidenceofmissingdataestimationforhivclassification |
_version_ |
1719084076470632448 |