Machine learning approaches classify clinical malaria outcomes based on haematological parameters

Table 3 Performance of classification models for identifying parameters that can be classified with clinical malaria

ANN	UM vs SM vs nMI	UM vs nMI	SM vs nMI
Model type	Multi-classification model	Binary model	Binary model
Data splitting
Total data (100%)	n = 2,207	n = 1,681	n = 1,504
Training and validation data (80%)	n = 1,766	n = 1,345	n = 1,204
Testing data (20%)	n = 441	n = 336	n = 300
Training performance
Training accuracy	0.862	0.856	0.985
Training loss	0.396	0.425	0.062
Validation accuracy	0.828	0.842	0.978
Validation loss	0.432	0.434	0.102
Testing performance
Testing accuracy	0.853	0.801	0.96
Kappa	0.768	0.583	0.913
ROC_AUC	NA^a	0.866	0.983
Precision	0.855	0.780	0.971
Recall	0.856	0.717	0.918
F1 score	0.856	0.747	0.944

Training and cross-validation accuracy as well as testing accuracy, area under the ROC curve (AUC), precision, recall, and F1 score. Multiclass analysis among all three-disease conditions, training accuracy was 0.862 with 0.828 validation accuracy. The model classified the three classes with 0.853 test accuracy. The ANN (UM vs nMI) had an accuracy of ≥ 0.801 for training, validation, and testing accuracy. The ANN (SM vs nMI) had the highest classification accuracy of ≥ 0.96
^aWe did not generate ROC-AUC for multi-classification models

ISSN: 1741-7015