Malaria remains a significant public health challenge and major cause of mortality among children in Nigeria. Children under five years are more vulnerable and significantly contribute to malaria-reported cases and deaths. However, factors leading to high malaria cases and deaths among children are not well studied and also, there is a lack of malaria predictive tools. Consequently, this study implemented machine learning approaches to identify factors influencing and predict malaria among Nigeria’s children under five years using the 2021 nationally representative Nigeria Malaria Indicator Survey (MIS) data. The study applied SMOTE sampling technique to handle class imbalance problem and XGBoost to generate feature importance scores. The study revealed that region, type of place of residence, religion, number of children under five in the household, educational attainment, household head’s sex, wealth index, type of mosquito bed net(s) slept under last night, birth order number are significantly associated with malaria prevalence in Nigeria’s under-fives. The study revealed that random forest achieved the highest accuracy score of 0.7898, recall of 0.7828, F1-score of 0.7883, precision of 0.7938, and AUC of 0.79. CatBoost lags behind the random forest with an accuracy of 0.7652, recall of 0.6517, F1-score of 0.7351, precision of 0.8430 and AUC of 0.77. Malaria predictive models can assist decision-makers in identifying factors influencing malaria prevalence, predicting malaria, developing targeted interventions and malaria data-driven tools, and identifying specific regions with a higher malaria transmission risk among children under five years.