Stability and Transformation Prediction of HIV Viral Load Using the Application of Machine Learning and Principal Component Algorithms

The current antiretroviral therapy (ART) medication gives hope to people living with human immunodeficiency virus (HIV) and those with acquired immunodeficiency syndrome (AIDS). This great development contributes to and improves the lives of HIV-treated patients. Lacking education and awareness among pupils in Nigeria and Africa about HIV is the reason we are suffering from an increasing number of new cases of HIV/AIDS. This study judges the ability of Machine Learning (ML) algorithms of Multilayer perceptron (MLP), Neuro fuzzy (NF), and Principal Component Analysis (PCA) to predict a retrospective cohort study of ART-treated patients at the Federal Teaching Hospital, Gombe (FTHG), State, Nigeria. The modelling used data from 1500 patients, including variables such as ART drug, hospital status, BP, viral load, age on ART, and birth and age groups. The ML models employed the (“coefficient of determination_DC and mean square error_MSE”) to measure the model accuracy, however, factor loading, the rotated component matrix, and the scree plot (variance and cumulative) employed PCA to evaluate the dimensionality of the components. Before the real prediction, the data underwent thorough proofed of (training 75%) and testing 25%). The NF demonstrated superior performance, achieving (91% and 90%) accuracy in both training and testing over MLP (90% and 89%). By applying PCA to simplify the high-dimensional data, three significant components were identified that met the Kaiser Criterion with eigenvalues of 1.946, 1.070, and 1.047. Together, these components effectively reduce the complexity of the original dataset while retaining more than 58% of the total information. The rotated component matrix showed that Component 1 was defined by a strong positive cluster of age group and patient age (both 0.985). Component 2 included positive loadings for hospital status (0.576) and SBP (0.726). Finally, Component 3 showed a mixed pattern, with a negative loading on viral load (-0.641) and a positive loading on ART start year (0.699). With an accuracy of 91%, the AI model confirms that years of ART treatment align with the 90-90-90 HIV/AIDS targets. Additionally, since the PCA components preserve over 58% of the original variance, this approach proves to be a robust method for clinical predictions.