Abstract:
Infection with the human immunodeficiency virus and acquired immunodeficiency syn drome (HIV/AIDS) continue to pose a threat to Amhara society. The number of people
who are aware of their HIV status has increased thanks to a variety of strategies. The most
accurate approach of counting the number of HIV contacts who might be at risk of con tracting HIV from HIV-positive people among these methods is index case testing. How ever, the manual nature of the current HIV index case testing has a number of challenges,
time-consuming and costly to operate. In order to forecast and depict HIV index case test ing, this study presents the findings of the machine-learning model. The process used to
construct the model adhered to the principles of agile software development. The data,
consisting of 15 characteristics and 14922 samples, was collected in the Amhara regions
from CommCare and SmartCare. The dataset was then divided into testing sets of 2984
samples (80/20) and training sets with 11938 samples each. The conventional Algorithms
which were applied Random Forest (RF), XGBoost, and Artificial Neural Networks
(ANN). The three conventional algorithms results were Random Forest accuracy (85%),
XGBoost accuracy (83.89%), and ANN accuracy (81.2%). The accuracy of random forest
algorithm was the highest. As a result, when compared to the other two algorithms, RF
appears to have the best performance. Data visualization shows that 55% of sexual partners,
23% of children, and 22% of spouses were among the index case contacts who received
test and received their test results. Our understanding of the value of machine learning in
predicting and visualizing HIV index case tests has improved as a result of our work. The
developed model can aid in the development of a workable intervention that will prevent
the spread of HIV and AIDS in our communities. According to the paper, multiple health
centers apply this idea to improve the way they operate.
Key terms: HIV, Index Case testing, Machine-leaning, Random Forest, XGBoost, Arti ficial Neural Network, accuracy