Publicação
A Predictive Model for Malaria in Pregnancy: Developing a predictive model to identify pregnant womenat highrisk of malaria infections
| Resumo: | Despite a massive amount of research that has been done to apply Machine Learning (ML) algorithms in predicting malaria disease, there is a significant gap in the literature concerning pregnant women, a vulnerable group, as a focus group. This study seeks to address this gap by proposing an approach to predictive modelling specifically for pregnant women at risk of malaria. We evaluated the most well-known ML classification algorithms using a dataset comprising 2,241 observations containing clinical, demographic, and laboratory results of consented participants attending antenatal clinic visits from October 2019 to May 2023 in settings belonging to five sub-Saharan countries. Our findings underscore the efficacy of ensemble methods, particularly Gradient Boosting and Random Forest Classifier, which exhibited a better performance than other classifiers with an F1-score of 90.3% (95% CI: 88 – 92) and 89.9% (95% CI: 86 – 90) respectively. Notably, in the Multi-Layer Perceptron classifier, trained with different scalers, minor differences were found in their performance, when applying Robust scaler the classifier achieved an F1-score of 89.0% (95% CI: 86 – 90) and the Standard Scaler F1-score was 86.9% (95% CI: 84 – 89). Moreover, in this experiment, the Adaboost F1-score was 83.9% (95% CI: 82 – 87) and the DT F1-score was 87.2 (95% CI: 84 – 99) where entropy was chosen as a function to measure the quality of a split in DT. Lastly, KNN and LR classifiers scored 82.7% (95% CI: 81 – 86) and 67.4% (95% CI: 65 - 69). Those findings could have profound implications for healthcare strategies targeting malaria prevention in pregnant women, offering a nuanced understanding of the efficacy of predictive modelling within this cohort. By leveraging ML algorithms, tailored interventions can be developed to mitigate malaria risks during pregnancy, thereby advancing maternal and child health outcomes in endemic regions. |
|---|---|
| Autores principais: | Tchavana, Corssino Jaime |
| Assunto: | Malaria Pregnancy Machine Learning Predictive modelling SDG 3 - Good health and well-being |
| Ano: | 2024 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso embargado |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | Despite a massive amount of research that has been done to apply Machine Learning (ML) algorithms in predicting malaria disease, there is a significant gap in the literature concerning pregnant women, a vulnerable group, as a focus group. This study seeks to address this gap by proposing an approach to predictive modelling specifically for pregnant women at risk of malaria. We evaluated the most well-known ML classification algorithms using a dataset comprising 2,241 observations containing clinical, demographic, and laboratory results of consented participants attending antenatal clinic visits from October 2019 to May 2023 in settings belonging to five sub-Saharan countries. Our findings underscore the efficacy of ensemble methods, particularly Gradient Boosting and Random Forest Classifier, which exhibited a better performance than other classifiers with an F1-score of 90.3% (95% CI: 88 – 92) and 89.9% (95% CI: 86 – 90) respectively. Notably, in the Multi-Layer Perceptron classifier, trained with different scalers, minor differences were found in their performance, when applying Robust scaler the classifier achieved an F1-score of 89.0% (95% CI: 86 – 90) and the Standard Scaler F1-score was 86.9% (95% CI: 84 – 89). Moreover, in this experiment, the Adaboost F1-score was 83.9% (95% CI: 82 – 87) and the DT F1-score was 87.2 (95% CI: 84 – 99) where entropy was chosen as a function to measure the quality of a split in DT. Lastly, KNN and LR classifiers scored 82.7% (95% CI: 81 – 86) and 67.4% (95% CI: 65 - 69). Those findings could have profound implications for healthcare strategies targeting malaria prevention in pregnant women, offering a nuanced understanding of the efficacy of predictive modelling within this cohort. By leveraging ML algorithms, tailored interventions can be developed to mitigate malaria risks during pregnancy, thereby advancing maternal and child health outcomes in endemic regions. |
|---|