Yusrina binti Mohd Yusoff1*, Najjah Tohar1, Suraini Mohd Suhaimi1, Mohd Azahadi Omar1, Mohd Zulfadli Hafiz Ismail1
1. Sector for Biostatistics & Data Repository, National Institutes of Health (NIH), Ministry of Health, Malaysia
*Corresponding author: Yusrina binti Mohd Yusoff.
ABSTRACT
INTRODUCTION: Anaemia is a common health issue affecting many women, particularly those of reproductive age. Early identification of individuals at risk through predictive modelling can enhance prevention strategies.
OBJECTIVE(S): This study aimed to assess the prevalence of anaemia and compare the performance of logistic regression and two machine learning models (Random Forest and XGBoost) in predicting anaemia risk. It also aimed to develop a predictive model using decision tree-based machine learning techniques with data from NHMS 2019.
MATERIALS & METHODS: A cross-sectional analysis was conducted using data from 8,492 female respondents with complete information on anaemia status and relevant variables. Six variables were included in all models: age, body mass index (BMI), diabetes, cholesterol, pregnancy status and fruit intake. Analyses were conducted using the same set of variables across both machine learning models and the logistic regression model. Machine learning models were trained and tested using a 70/30 train-test split. Model performance was evaluated using accuracy, sensitivity, specificity and the Area Under the Receiver Operating Characteristic Curve (AUC).
RESULTS: Anaemia prevalence was 23% with key predictors including increasing age, low BMI, diabetes and high cholesterol. Random Forest achieved 78.9% accuracy and an AUC of 0.864, while XGBoost achieved 80.2% accuracy and an AUC of 0.846. Random Forest offered better sensitivity (93.5%) and overall model balance. Logistic regression showed similar predictors as statistically significant (p < 0.05), while pregnancy and fruit intake showed borderline significance. Consistency across models reinforces the robustness of these findings.
CONCLUSION: Machine learning models, particularly Random Forest, demonstrate strong potential for identifying anaemia risk and support the use of hybrid approaches to improve women’s health interventions.
KEYWORDS: Anaemia, Machine Learning, NHMS 2019, Women’s Health, Logistic Regression


