A MACHINE LEARNING MODEL FOR PREDICTING PRE ECLAMPSIA FOR RESOURCE-CONSTRAINED REGIONS.

Abstract

Pre-eclampsia is globally recognized by the World Health Organization as a significant contributor to high rates of morbidity and mortality among infants and mothers. It accounts for approximately 3% to 5% of all reported pregnancy-related complications worldwide. However, in developing nations like Kenya, particularly in the sub-Saharan region, the prevalence of pre-eclampsia is notably higher, ranging from 5.6% to 6.5% of reported pregnancies. Key risk factors associated with pre-eclampsia include sudden elevation in blood pressure, increased protein levels in urine, chronic kidney disease, and the presence of either Type 1 or Type 2 diabetes. This research developed a predictive model for pre-eclampsia utilizing supervised machine learning techniques on socio-demographic data gathered from Kilifi County. A total of 500 secondary data records gathered from Kilifi county hospital were pre-proposed to train and test the machine learning models. To train and test the models, the study employed five supervised machine learning algorithms, namely Logistic Regression, Random Forest, Naïve Bayes, Linear Discriminant Analysis, and Support Vector Machines. Maternal age, marital status, gravida, education level, and ANC attendance were identified as the optimal extracted features using PCA. The logistic regression model outperformed other supervised machine learning models in the study, achieving a high accuracy rate of 0.96 in predicting pre-eclampsia. The results show that Logistic Regression can accurately predict pre-eclampsia within the first trimester of pregnancy. Future research will involve collecting more data from different regions to improve performance and building a mobile application that will improve MCH accessibility in resource-constrained regions in the country