Neural Network Modelling in HIV/AIDS –Five Years Survival Cohort Data

Neural Network is very good area to deal with most of the medical problems. It has many algorithms for classification, prediction, image processing etc., The main objective of this research is to find out the application of Neural Network technique to classify and provide solutions and to improve methodological aspects in ART treatment. A total of 110 PLHIV on HAART-A five years cohort data (from 2004) was obtained from CMIS software back up to construct different neural net work models like Back Propagation Neural Network algorithm, Network and Radial Basis Function Network to predict the mortality and efficacy of the ART treatment. As per the hierarchial neural net work the actual predicted mortality rate at the end of the five year cohort was 26.30%, relative error (RE) 0.994.Specificity and sensitivity of the model 93.26%, & 84.28% respectively and the accuracy of the model was tested using PPV (Positive predictive value) and Negative predictive value (NPV) .The PPV expressed was 90.26% NPV rate 9.74%. Nonhierarchial models took approximately more epochs than hierarchial . Actual predicted mortality rate was 18.63% .The specificity, or true negative rate, of the non hierarchial model at this level was less superior (Specificity83.21%, Sensitivity -76.28%,PPV-64.36% with actual rate of NPV -25.64%). the pattern of frequencies has differed in both the model. Comparison of twoneural network models for survival analysis is done. For a cohort of 110 PLHIV , the hierarchial neural-network models for survival analysis could provide efficient patterns faster than could a nonhierarchial model. The hierarchial models also provide greater accuracy and more reliability in predicting mortality rates.