A Comparative Study of Selected Supervised Machine Learning Algorithms for Detecting Child Mortality
The World Health Organization (WHO) defines under-five mortality rate as the probability that a child born in a specified year will die before reaching the age of five. It is a key metric for determining the quality of a country's healthcare system.In many parts of the world, under-five mortality is still a big concern, according to the World Health Organization (WHO). Under-five mortality has been a major issue in emerging nations. Even though the number of childhood deaths worldwide has significantly decreased, under-five mortality is still a significant public health concern in Nigeria. Nigeria now has the second-highest number of under-five deaths globally, after India, and the highest recorded number of under-five deaths in Africa (WHO, 2022). In Nigeria, around one million children under the age of five pass away each year, with over 60% of these deaths taking place between the ages of one and fifty-nine. According to data from the Nigeria Demographic and Health Surveys (NDHS), the under-five mortality rate increased from 157 per 1000 live births in 2008 to 138.8 per 1000 live births in 2020. This showed that the rate is still high, and the Millennium Development Goal (MDG) calls for reducing under-five mortality to 25 per 1,000 live births by 2030. The study's precise goals are to use supervised machine learning classification algorithms to predict mortality among children under five using nationally representative data on children under five and the specific objectives are, to predict under five mortality using machine learning classification models,to compare the performance of the supervised machine learning Classification models and lastly to apply the study's conclusions to improve and enhance relevant government policies and intervention plans. The study utilized secondary data from the 2018 Nigeria demographic and health survey data (NDHS). The 2018 NDHS is a nationally probability sample survey with approximately 42,000 households (NDHS 2018). The unit of analysis include children under-five years with total sample size of 33924 selected from 1389 clusters across Nigeria. The (2018 NDHS) dataset was downloaded from www.dhsprogram.com website, SPSS version 23 was used to open the data set and R programming Language is used to perform data analysis and processing. In this study, the outcome” variable was under-five “mortality measured as a binary outcome, under-five mortality measured as being alive (coded as 0) or dead (coded as 1)”.Twenty seven(27)predictors (features) used in this study are mother’s age at birth, highest educational level, mother’s body mass index (BMI), Currently breastfeeding, Child alive, whether the child was wanted, sex of the child, birth order, child is twin, age in five years, given child anything other than breast, Birth weight in kg, Current age of child, Size of child at birth, source of drinking water, toilet facility, household wealth index, marital status, place of residence type, region, Native language ,religion, antenatal care visit, Delivery by caesarean, place of delivery, covered by health insurance and Number of antenatal care visit. The study used 2018 demographic and health survey data to investigate the factors linked to under-five mortality in Nigeria. Four distinct classification techniques for supervised machine learning were compare. Support Vector Machine (SVM) and Naïve Bayes emerged as the top performers among the four evaluated models based on their accuracy followed by Artificial Neural Network and lastly Linear Discriminant Analysis(LDA) models with accuracy values reaching 62.48%. This indicates their strong ability to correctly identify cases where a child is not alive while maintaining a moderate level of sensitivity 62.74% for detecting cases where a child is alive and Lastly Support vector machine model outperformed all other models by achieving perfect scores across all metrics, suggesting that it can accurately classify every instance in the dataset.
Download Paper