Machine Learning Based Detection of Chronic Kidney Disease using Full Blood Count from a Ghanaian Hospital Population with an Integrated Web Application

Random Forest and Extra Trees trained on ten full blood count features detected chronic kidney disease in 155 Ho Technical University Hospital patients with 83.9% accuracy and AUC up to 0.939.

• Why it matters: CKD is often diagnosed late in sub-Saharan Africa because confirmatory tests are frequently unaffordable, while full blood count testing is inexpensive and widely available.
• What they did: Using repeated stratified cross-validation, 11 algorithms, six resampling strategies, and 13 ensembles were compared, with synthetic minority oversampling (SMOTE) applied either before or after the train-test split.
• The result: When SMOTE was applied before the split, accuracy was 83.9% (AUC 0.925–0.939), but restricting resampling to training folds reduced held-out performance to 61.5% accuracy and AUC 0.604 in 39 unseen patients.

AI summary of the abstract · may contain errors · not medical advice · read the paper.