Malaria remains a critical global health challenge requiring innovative diagnostic approaches. Machine learning offers promising solutions for automated detection, but systematic algorithm comparison using clinically validated data remains limited. We systematically compared five machine learning models—Naive Bayes, Logistic Regression, Random Forest, XGBoost, and Enhanced Bayesian Logistic Regression—for malaria detection using a rigorously validated synthetic dataset ( \(N=10,100\) ) representing Sub-Saharan African epidemiological conditions. The dataset achieved 87% representativeness against published clinical benchmarks. Cost-sensitive threshold optimization prioritized clinical sensitivity ( \(C_{FN}=15\) , \(C_{FP}=3\) ). Performance evaluation employed comprehensive metrics with bootstrap confidence intervals, statistical significance testing, and clinical cost analysis. Five machine learning algorithms were evaluated using comprehensive statistical validation including bootstrap confidence intervals and significance testing. XGBoost achieved optimal performance with highest \(\text {AUC}\) (0.956, 95% \(\text {CI}\) : 0.952–0.961) and competitive clinical cost (5,496), representing 2.8% improvement over Random Forest. Enhanced Bayesian Logistic Regression incorporating clinical domain knowledge achieved comparable performance ( \(\text {AUC}\) : 0.954, 95% \(\text {CI}\) : 0.950–0.959) with interpretable clinical coefficients. McNemar’s test revealed statistically significant classification differences between XGBoost and Random Forest ( \(\chi ^2 = 1508.6\) , \(p < 0.001\) ), while Friedman test indicated no significant overall ranking differences across models ( \(\chi ^2 = 4.0\) , \(df = 4\) , \(p = 0.406\) ). Complete analysis pipeline with 2-step reproduction process and synthetic datasets are publicly available for independent validation and collaborative research extension. XGBoost demonstrates optimal balance of accuracy and cost-effectiveness for malaria screening applications. The systematic validation framework and cost-sensitive optimization provide practical guidance for clinical implementation. While synthetic data enables controlled algorithm comparison, real-world clinical validation remains essential before deployment.
