Developing the Predictive Model for the Level of Food Insecurity Status of Households Using Ensemble Machine Learning Techniques with XAI

Among the world’s undernourished people, more than 282 million people in Africa and 418 million people in Asia were found. Moderate or severe food insecurity at the global level has risen from 22.6% in 2014 to 26.6% in 2019, with Sub-Saharan Africa experiencing high levels, particularly in Ethiopia. In 2018, 239 million undernourished people were registered in Sub-Saharan Africa particularly high in Kenya, Somalia, Ethiopia, and South Sudan. This study, hence, aimed to develop a predictive model using ensemble machine learning algorithms to assess and classify the levels of food insecurity based on socioeconomic, environmental, and demographic factors in North Western Ethiopia. The dataset, collected from the Dabat Health and Demographic Surveillance from 2014 to 2020, was preprocessed using data cleaning, transformation, and class balancing with SMOTE-ENN. The dataset is split into training, validation, and testing sets with a 90/10 ratio. Relevant features were selected via recursive feature elimination, and class decomposition methods were evaluated. Experiments with ensemble models, including Random Forest, Gradient Boosting, XGBoost, CatBoost, and LightGBM, identified XGBoost as the best-performing algorithm, achieving 91.53% accuracy, 89.39% recall, and a 94.00% micro-average ROC score. SHAP was employed for explainable AI, offering insights into critical factors influencing food insecurity. Key factors that enhance food security include receiving assistance in the form of food or money, the size of cultivated land, altitude, monthly income, and household assets. In contrast, factors that increase the likelihood of food insecurity include the education level of the household head, age, marital status, limited access to irrigation, and reductions in portion sizes and meal frequency for the household head. This study demonstrates the utility of machine learning in identifying key drivers of food insecurity and offers a robust framework for targeted interventions. By bridging predictive analytics with explainable AI, actionable insights are provided to policymakers and stakeholders to alleviate food insecurity in Ethiopia and beyond.