Evaluation of Artificial Neural Network Performance in Classifying Anemia Cases Among Pregnant Women in Benghazi, Libya

Abstract

This study aimed to evaluate the ability of artificial neural networks (ANNs) to classify anemia cases among pregnant women, as anemia represents a major health challenge affecting the health of both mothers and fetuses. The study also sought to identify the most important variables associated with the classification process. The significance of this research lies in the potential use of modern classification methods to analyze health data and identify patterns and relationships among different variables. The study was based on secondary data collected from a sample of 600 pregnant women in Benghazi, Libya. The participants were randomly selected from several clinics and hospitals during the period 2020–2021, as reported in Nasr (2023). The dataset included a variety of demographic, health-related, and pregnancy-related variables. Anemia status was determined based on blood hemoglobin levels. Cases with hemoglobin levels below 11 (g/dL) were classified as anemic, whereas cases with levels of 11 (g/dL) or higher were classified as non-anemic, according to the criterion adopted in the study. Artificial neural networks were used to construct a classification model. The data were divided into two sets: a training set and a testing set, in order to evaluate the model's ability to classify cases correctly. The results showed that the classification accuracy reached 90.8% for the training data and 88.8% for the testing data. These findings indicate that the model achieved a good level of classification performance, with relatively similar results across the training and testing sets. The variable importance analysis also showed that bleeding during pregnancy ranked first, followed by nutritional supplements, iron tablet intake, previous miscarriage, weight, age, menstrual cycle duration, educational level, and gestational age in weeks. These findings highlight the potential of artificial neural networks for classifying health data and identifying the variables that contribute most to the classification process. The study recommends conducting further research using independent samples to assess the model's generalizability and utilizing classification results to support health research on anemia among pregnant women. It also emphasizes that the importance of variables within the model does not necessarily imply a causal relationship between those variables and the occurrence
of anemia.