Evaluating the Effectiveness of Zero-Inflated Models and Machine Learning Techniques for Malaria Prevalence: A Comparative Study

Abstract

In this study, Malaria is still a serious global health issue that demands new approaches for precise prevalence estimation and efficient control measures. This study undertakes a comprehensive investigation into the evaluation of statistical models and machine learning techniques for estimating malaria prevalence in regions where zero counts are prevalent. The research focuses on zero-inflated data typical of malaria prevalence and seeks to identify the most appropriate modelling strategy for precise prevalence estimation in such circumstances. The research question driving this study is: ”Which zero-inflated model exhibits superior performance in handling zero-inflated malaria data?” To address this question, the study engages in a comparative analysis of various modeling techniques. Zero-inflated models including Zero-Inflated Poisson (ZIP) and Zero-Inflated Negative Binomial (ZINB) are employed alongside traditional Generalized Linear Models (GLM) such as Poisson and Negative Binomial (NB) models. Additionally, hurdle models including Poisson Hurdle and Negative Binomial Hurdle models are incorporated into the analysis to account for the excess zeros in the data. The study uses a comprehensive approach that involves fitting and assessing these various models to actual data on the prevalence of malaria. The primary focus was to assess the efficacy of these models in accurately capturing the characteristics of the data. Our findings revealed varying degrees of model performance. Notably, the ZINB and ZIP models exhibited limitations in effectively handling the complexity of the zero-inflated data, resulting in suboptimal fits. However, contrasting results emerged when employing the negative binomial (NB) negative binomial hurdle (NBH), poisson (P) and poisson hurdle (HP) models, as they consistently demonstrated robust performance across different aspects of the data. These results underscore the importance of carefully selecting an appropriate modeling approach for zero-inflated data analysis. The success of the NB, NBH, P, HP models in our study suggests their potential as effective tools for modeling zero-inflated data with greater accuracy and reliability. This work contributes to the understanding of model selection in the context vii of zero-inflated data, offering valuable insights for researchers and practitioners working with similar data types. Moreover, advance machine learning algorithms such as support vector machine and random forest model also effectively captured zero inflated data