Explainable Machine Learning for Multiclass Classification of Concurrent Child Undernutrition in Ethiopia

Maternal and child undernutrition remains a major public health challenge globally, with the highest burdens observed in low- and middle-income countries. Coexisting maternal and child undernutrition has serious implications for survival, growth, and quality of life.Maternal and child undernutrition is a complex, multifactorial issue influenced by a web of interconnected determinants including health, socio-economic status, education and environmental conditions. To examine the spatial distribution and multilevel determinants of coexisting maternal and child undernutrition in Ethiopia using EDHS data from 2000-2016. We analyzed a weighted sample of 33,445 participants from four consecutive EDHS surveys. Spatial autocorrelation, hotspot, and interpolation analyses were conducted using ArcMap 10.8. Multilevel logistic regression was performed in Stata 17. Cluster variability was assessed using ICC, MOR, and PCV. Variables with p < 0.05 were considered statistically significant. A total of 33,445 weighted sample was used for this study. The prevalence of coexisting maternal and child undernutrition was 22.87% (95% CI: 22.42, 23.32). Spatial analysis result showed that hot spot areas were concentrated in the northern regions especially Tigray, Amhara, and parts of Benishangul-Gumuz. Multilevel logistics regression analysis result revealed that maternal primary education (AOR = 0.88; 95% CI: 0.78, 0.98), secondary/higher education (AOR = 0.38; 95% CI: 0.29, 0.49), medium household wealth (AOR = 0.85; 95% CI: 0.75, 0.97), high household wealth (AOR = 0.78; 95% CI: 0.69, 0.89), child age 12-23 months (AOR = 2.77; 95% CI: 2.44, 3.16), ANC use (AOR = 0.83; 95% CI: 0.75, 0.92), improved toilet (AOR = 0.83; 95% CI: 0.71, 0.98) and regions. Coexisting maternal and child undernutrition shows marked geographic clustering in northern Ethiopia. Strengthening maternal education, improving household economic conditions, enhancing ANC use, and expanding sanitation services particularly in high-risk regions are essential to reduce the burden.

This paper looks at how changing food prices affect child undernutrition in Ethiopia. It derives height for age (stunting) and weight for height (wasting) as indicators of child undernutrition from the two most recent years of the Livings Standards Measurement Survey and utilizes market prices for key cereals, teff, wheat, and maize at the zone level across all regions of the country. Using a panel data fixed effects model, the analysis finds that, contrary to previous studies, rising crop prices are positively associated with improved child stunting rates for children between ages 6 months and 5 years, while the results for wasting are not conclusive. These results suggest that across the board policy interventions that seek to suppress cereal price increases may have adverse effects on poverty reduction in the long term by undermining potentially positive impacts on child nutrition.

IntroductionDynamic Bayesian networks improve the modeling of complex systems by incorporating continuous probabilistic relationships between covariates that change over time. This study aimed to analyze the complex causal links contributing to child undernutrition using dynamic Bayesian network modeling, examining both the best- and worst-case scenarios. The Young Cohort of the Ethiopian Young Lives dataset from 2002–2016 was used to analyze the complex relationships among various covariates influencing child undernutrition. We used a built-in Bayes server tool to identify potential features, followed by building the structure of the directed acyclic graph using a structural learning algorithm. The maximum posterior is determined using the relevance tree algorithm. The node with the highest values of mutual information and target entropy reduction, along with the lowest value of target entropy, was considered to have the strongest predictive power in the dataset.ResultsThis study revealed that long-term participation in programs increased the likelihood of children being in a normal nutritional state. Key factors influencing the nutritional status of children under two years of age include the mother’s education level, her subjective well-being, and the household’s wealth quintile. Children with educated parents were more likely to have a healthy nutritional status. Additionally, the causal pathway of intervention programs → wealth quintile → child nutritional status consistently exceeded 90% in Waves 3, 4, and 5, indicating a strong relationship. Similarly, the relationship between intervention programs → food security → child nutritional status was nearly perfect at 99.99% in Waves 4 and 5, indicating a strong association. Finally, the study revealed that household participation in intervention programs significantly reduces undernutrition in best-case scenarios, while the absence of support poses a higher risk in worst-case conditions.ConclusionThe comprehensive intervention program strongly improved household wealth, food security, and maternal well-being, which in turn affected children’s nutritional status.

Stunted from the start: Early life weather conditions and child undernutrition in Ethiopia.

To examine the contribution of child, maternal and household factors in stunting, wasting and underweight among children under 5 years in Ethiopia. Quantitative cross-sectional design based on nationally representative data. Urban and rural areas of Ethiopia.ParticipantsYounger (0-24 months; n 4199) and older age groups (25-59 months; n 5497), giving a total of 9696 children. Among the younger age group, 29 % were stunted, 14 % were wasted and 19 % were underweight; and among the older age group, the prevalence of stunting, wasting and underweight was 47, 8 and 28 %, respectively. Being female, intake of multiple micronutrients, household having a piped source of drinking-water, high maternal BMI, higher household wealth and higher maternal education were associated with decreased odds of at least one form of undernutrition in both groups. On the other hand, children who were anaemic, had low birth weight, drank from a bottle, and children of stunted or wasted or working mothers were more likely to be stunted, wasted or underweight in both groups (P&lt;0·05). While most predictors and/or risk factors followed a similar pattern across the two age groups, child factors had higher leverage in the younger than the older group across the three forms of undernutrition. Multiple set of factors predicted childhood undernutrition in Ethiopia. The study underscores the importance of intervening in the first 1000 days through promoting maternal education, maternal-child health services, mother's nutrition and improving intrahousehold food distribution.

The prevalence of under-five children's undernutrition in Ethiopia is among the highest in the world. This study aimed at exploring the prevalence and risk factors of the composite index for anthropometric failure (CIAF) of under-five children in Ethiopia by incorporating the zonal (district) effects. The data was drawn from Ethiopian Demographic and Health Surveys (EDHSs), a population-based cross-sectional study of 29,599 under-five year children from 72 Zones in the years 2000, 2005, 2011, and 2016. Fixed effect variables related to child and maternal-household were included in the model. We adopted a generalized mixed model with CIAF as outcome variable and Zones as random effects. The prevalence of CIAF in Ethiopia was 53.78% with the highest prevalence of 61.30% in 2000 and the lowest prevalence of 46.58% in 2016. The model result revealed that being a female child, absence of comorbidity, singleton births, and the first order of birth showed significantly lower CIAF prevalence than their counterparts. Among the household characteristics, children from mothers of underweight body mass index, uneducated parents, poor household sanitation, and rural residents were more likely to be undernourished than their counterparts. Based on the best linear unbiased prediction for the zonal-level random effect, significant variations of CIAF among zones were observed. The generalized linear mixed-effects model results identified gender of the child, size of child at birth, dietary diversity, birth type, place of residence, age of the child, parental level of education, wealth index, sanitation facilities, and media exposure as main drivers of CIAF. Disparities of CIAF were observed between and within the Ethiopian administrative Zones over time.

As Federated Learning (FL) gains prominence in distributed machine learning applications, achieving fairness without compromising predictive performance becomes paramount. The data being gathered from distributed clients in an FL environment often leads to class imbalance. In such scenarios, balanced accuracy rather than accuracy is the true representation of model performance. However, most state-of-the-art fair FL methods report accuracy as the measure of performance, which can lead to misguided interpretations of the model's effectiveness to mitigate discrimination. To the best of our knowledge, this work presents the first attempt towards achieving Pareto-optimal trade-offs between balanced accuracy and fairness in a federated environment (FairTrade). By utilizing multi-objective optimization, the framework negotiates the intricate balance between model's balanced accuracy and fairness. The framework's agnostic design adeptly accommodates both statistical and causal fairness notions, ensuring its adaptability across diverse FL contexts. We provide empirical evidence of our framework's efficacy through extensive experiments on five real-world datasets and comparisons with six baselines. The empirical results underscore the potential of our framework in improving the trade-off between fairness and balanced accuracy in FL applications.

Antiretroviral therapy (ART) adherence is crucial for health outcomes of people living with HIV (PLHIV), influenced by a complex interplay of individual, community, and household factors. This article focuses on the influence of household factors, as well as individual and community factors, on ART adherence among PLHIV in Cape Town who have recently initiated ART. Baseline data for a cluster-randomized controlled trial were collected from 316 PLHIV in 12 districts in Cape Town between 6th May 2021 and 22nd May 2022. Zero-inflated Poisson models, with cluster-adjusted standard errors, were used to analyse the association between individual, household, and community factors and ART adherence measures. At the household-level, household support was associated with both better self-rated adherence (exp(β) = 0.81, z = − 4.68, p < 0.001) and fewer days when pills were missed (exp(β) = 0.65, z = − 2.92, p = 0.003). Psychological violence (exp(β) = 1.37, z = 1.97, p = 0.05) and higher household asset scores (exp(β) = 1.29, z = − 2.83, p = 0.05) were weakly associated with poorer ART adherence. At the individual-level, male gender (exp(β) = 1.37, z = 3.95, p < 0.001) and reinitiating ART (exp(β) = 1.35, z = 3.64, p < 0.001) were associated with worse self-rated ART adherence; higher education levels (exp(β) = 0.30 times, z = − 3.75, p < 0.001) and better HIV knowledge (exp(β) = 0.28, z = − 2.83, p = 0.005) were associated with fewer days where pills were missed. At the community-level, community stigma was associated with worse self-rated ART adherence (exp(β) = 1.24, z = 3.01, p = 0.003). When designing interventions to improve ART adherence, household, individual and community factors should all be considered, particularly in addressing gender-based disparities, reducing stigma, tackling violence, and enhancing household support.Clinical Trial Number: Pan African Clinical Trial Registry, PACTR201906476052236. Registered on 24 June 2019.

Objective: To use SMOTE to enhance class balance and compare the performance of different classification methods before and after applying SMOTE. Methods: The study used a dataset from Kaggle. Consisted of several health-related features linked to obesity prediction. Checking for class imbalance within the dataset affected initial model performance. SMOTE was applied to synthetically increase the representation of minority classes, reducing the class imbalance. It was conducted in two stages: 1. Training and testing the classification algorithms before applying SMOTE. 2. Training and testing the same models after applying SMOTE to enhance class balance. The performance of all models was evaluated based on metrics before and after the SMOTE application. Results: Models Logistic Regression and Naive Bayes struggled with low sensitivity and specificity, and KNN (k=5) showed poor specificity. Significant improvements were observed across all models after applying SMOTE. Logistic Regression, despite a decrease in accuracy(-8.8), sensitivity and specificity increased substantially(+56.7%), with balanced accuracy improving(+16.6%). Naive Bayes saw a modest accuracy increase(+2.3%), with sensitivity and specificity improving(+47.9%). The KNN classifier exhibited a transformative enhancement with sensitivity and specificity increasing(+96.0%) and balanced accuracy(+28.3%). Deep Learning showed a significant increase in sensitivity (+69.8%), balanced accuracy (+29.4%), and an improvement in precision and F1-score despite a slight decrease in specificity(-10.9%). Conclusion: The results demonstrate that while there might be slight trade-offs, the overall improvements in key metrics such as sensitivity, specificity, balanced accuracy, precision, and F1-score affirm the utility of SMOTE in enhancing model performance for imbalanced datasets

Investigate the relationship between maternal autonomy at multiple levels and the risk of child stunting, underweight, and wasting in India. Data were from a 2005-2006 nationally representative, cross-sectional sample of 51,555 children under 5years from 29 states in India. Multilevel, multivariable, logistic regression analyses were used to estimate the odds of child stunting, underweight, and wasting in relation to maternal autonomy in healthcare, movement, and money at the individual level and community level, while adjusting for several child, maternal, and household factors. When only adjusting for child age and sex, children in communities with a high proportion of women with autonomy in healthcare, or movement, or money, separately, had a lower risk of being stunted, underweight, or wasted, separately. However, adjusting for other explanatory factors attenuated these relationships and made them statistically insignificant. Individual maternal autonomy in any of the three domains was not associated with any of the outcomes. The results suggest that caution should be taken when interpreting the direct relevance of maternal autonomy at both individual and community levels to measures of child undernutrition.

Purpose: Functional dependence affects well-being and life expectancy. We identified baseline variables that best predict future limitations in basic and instrumental activities of daily living. Method: Using a filtering approach, we selected the best predictors from 4,248 candidate predictors for 39,927 Canadian Longitudinal Study on Aging participants aged 44–88 years and compared machine learning models trained on baseline data (2010–2015) on their ability to classify functional status (independent vs. dependent people) at follow-up (2018–2021) in a training set ( n = 31,941). We then tested the best model in a separate test set ( n = 7,986). Results: Eighteen variables best predicted functional status at follow-up. Logistic regression performed best, achieving 81.9% balanced accuracy in the test set. Protective factors included no baseline limitations, stronger grip, absence of pain or chronic conditions, female sex, having a driver’s license, and good memory. Risk factors included older age, psychological distress, slow walking, retirement, chronic conditions, and physical inactivity. Conclusion: Functional status can be predicted up to 6 years in advance using health, demographic, cognitive, and physical activity variables. Early identification may enable timely interventions to delay functional decline.

This study evaluated how correct use of child protective equipment (child restraint systems, seat belts, and helmets) influences predicted injury severity for children involved in police-reported road crashes. Data from 69,108 participants under 18 years were analyzed, covering occupant, vehicle, roadway, environmental, and protection factors. An XGBoost classifier achieved ROC AUC = 0.8186 with balanced accuracy, precision, and recall. SHAP interpretation identified seating position and participant type as the most influential predictors. Counterfactual simulations, assuming full compliance with protective-equipment use, showed improved predicted outcomes in 64 cases, while 15 worsened. Helmet non-use was the most frequent lapse. Consistent, correct use of protective devices significantly shifts predicted outcomes toward less severe injuries. The explainable machine-learning and counterfactual framework quantifies the benefits of compliance and provides actionable evidence for targeted education, enforcement, and vehicle-safety design. The approach can be extended to other vulnerable groups, including pregnant occupants.

Background/Objectives: Fetal health is essential in prenatal care, influencing both maternal and fetal outcomes. Cardiotocography (CTG) monitors uterine contractions and fetal heart rate, yet manual interpretation exhibits significant inter-examiner variability. Machine learning offers automated alternatives; however, class imbalance in CTG datasets where pathological cases constitute less than 10% leads to poor detection of minority classes. This study aims to provide the first systematic benchmark comparing five resampling strategies across seven classifier families for multi-class CTG classification, evaluated using imbalance-aware metrics rather than overall accuracy alone. Methods: Seven machine learning models were employed: Naïve Bayes (NB), Random Forest (RF), Linear Discriminant Analysis (LDA), k-Nearest Neighbors (KNN), Linear Support Vector Machine (SVM), Multinomial Logistic Regression (MLR), and Multi-Layer Perceptron (MLP). To address class imbalance, we evaluated the original unbalanced dataset (base) and five resampling methods: SMOTE, BSMOTE, ADASYN, NearMiss, and SCUT. Performance was evaluated on a held-out test set using Balanced Accuracy (BACC), Macro-F1, the Macro-Matthews Correlation Coefficient (Macro-MCC), and Macro-Averaged ROC-AUC. We also report per-class ROC curves. Results: Among all models, RF proved most reliable. Training on the original distribution (base) yielded the highest BACC (0.9118), whereas RF combined with BSMOTE provided the strongest class-balanced performance (Macro-MCC = 0.8533, Macro-F1 = 0.9073) with a near-perfect ROC-AUC (approximately 0.986–0.989). Overall, resampling effects proved model dependent. While some classifiers achieved optimal performance on the natural class distribution, oversampling techniques, particularly SMOTE and BSMOTE, demonstrated significant improvements in minority class discrimination and class-balanced metrics across multiple model families. Notably, certain models benefited substantially from resampling, exhibiting enhanced Macro-F1, BACC, and minority class recall without sacrificing overall accuracy. Conclusions: These findings establish robust, model-agnostic baselines for CTG-based fetal health screening. They highlight that strategic oversampling can translate improved minority class discrimination into clinically meaningful performance gains, supporting deployment in cost-sensitive and threshold-aware clinical settings.

to predict vestibular schwannoma (VS) response to radiosurgery by applying machine learning (ML) algorithms on radiomic features extracted from pre-treatment magnetic resonance (MR) images. patients with VS treated with radiosurgery in two Centers from 2004 to 2016 were retrospectively evaluated. Brain T1-weighted contrast-enhanced MR images were acquired before and at 24 and 36 months after treatment. Clinical and treatment data were collected contextually. Treatment responses were assessed considering the VS volume variation based on pre- and post-radiosurgery MR images at both time points. Tumors were semi-automatically segmented and radiomic features were extracted. Four ML algorithms (Random Forest, Support Vector Machine, Neural Network, and extreme Gradient Boosting) were trained and tested for treatment response (i.e., increased or non-increased tumor volume) using nested cross-validation. For training, feature selection was performed using the Least Absolute Shrinkage and Selection Operator, and the selected features were used as input to separately build the four ML classification algorithms. To overcome class imbalance during training, Synthetic Minority Oversampling Technique was used. Finally, trained models were tested on the corresponding held out set of patients to evaluate balanced accuracy, sensitivity, and specificity. 108 patients treated with Cyberknife® were retrieved; an increased tumor volume was observed at 24 months in 12 patients, and at 36 months in another group of 12 patients. The Neural Network was the best predictive algorithm for response at 24 (balanced accuracy 73% ± 18%, specificity 85% ± 12%, sensitivity 60% ± 42%) and 36 months (balanced accuracy 65% ± 12%, specificity 83% ± 9%, sensitivity 47% ± 27%). radiomics may predict VS response to radiosurgery avoiding long-term follow-up as well as unnecessary treatment.

Objectives: Anemia continues to be a public health problem in India with an estimated 57% of women of reproductive age affected. The objective of the current study was to identify individual- and household-level factors associated with anemia among non-pregnant (NPWRA) in Punjab, India. Methods: The current analysis used baseline data from a trial of quintuply-fortified salt in Punjab, India. Hemoglobin concentrations measured from venous blood samples using an automated hematology analyzer (Sysmex XP-300, Kobe, Japan) were available for 798 NPWRA. Anemia was defined as a hemoglobin concentration < 12 g/dL. Bivariate associations between the participant’s demographic characteristics (age, education, number of children under five years, employment status), individual factors (dietary diversity, body mass index, and depressive symptoms), household factors (wealth, size, food security, education of household head), and anemia were first employed using Chi-square tests of independence. Variables that were significant at p < 0.1 in the bivariate analysis were included in the multivariate logistic regression model. Results: Mean ± SD age of NPWRA was 30.8 ± 5.6 years, 86% had at least a middle school education, and 52% had a BMI ≥ 25 kg/m2. The mean hemoglobin concentration was 11.9 ± 1.3 g/dL and 46% of NPWRA were anemic. In bivariate models, the following factors were associated with anemia: lower education, dietary diversity and household wealth, and higher household size. In the multivariate analysis, each variable remained associated (p < 0.1), and household size had the strongest association: households with < 6 members were 33% less likely to have anemia vs. households with ≥ 6 members (OR=0.66; 95% CI: 0.50, 0.88, p=0.005). Conclusions: NPWRA living in large households with low socioeconomic status may need intensified interventions to reduce anemia in this setting. Funding Sources: Bill & Melinda Gates Foundation (INV 002945).