A novel method for screening malignant hematological diseases by constructing an optimal machine learning model based on blood cell parameters

Table 2 The performance measures evaluated for different ML and AI methods for screening malignant haematological diseases use different thresholds found in the validation set

Model	Thresholds	AUC	Accuracy	Precision	Sensitivity	Specificity	F1-score
The thresholds of 0.5
LR	0.5	0.834	0.749	0.722	0.825	0.671	0.671
NB	0.5	0.836	0.709	0.897	0.485	0.942	0.942
KNN	0.5	0.809	0.759	0.834	0.657	0.864	0.864
SVM	0.5	0.881	0.808	0.804	0.825	0.791	0.791
RF	0.5	0.907	0.838	0.830	0.858	0.818	0.818
MLP	0.5	0.864	0.785	0.786	0.795	0.775	0.775
GBDT	0.5	0.913	0.831	0.823	0.851	0.810	0.810
ANN	0.5	0.906	0.857	0.881	0.832	0.884	0.884
The threshold for the best optimal ROC Youden index
LR	0.602	0.834	0.760	0.786	0.728	0.795	0.795
NB	0.000	0.836	0.760	0.848	0.646	0.880	0.880
KNN	1.000	0.809	0.490	/	0.000	1.000	1.000
SVM	0.646	0.881	0.821	0.869	0.765	0.880	0.880
RF	0.440	0.907	0.842	0.818	0.888	0.795	0.795
MLP	0.515	0.864	0.789	0.794	0.791	0.787	0.787
GBDT	0.614	0.913	0.840	0.857	0.825	0.857	0.857
ANN	0.522	0.906	0.857	0.891	0.821	0.895	0.895
The threshold for the best optimal precision-recall
LR	0.446	0.834	0.749	0.705	0.873	0.620	0.620
NB	0.000	0.836	0.757	0.713	0.873	0.636	0.636
KNN	0.500	0.809	0.759	0.834	0.657	0.864	0.864
SVM	0.615	0.881	0.821	0.851	0.787	0.857	0.857
RF	0.440	0.907	0.842	0.818	0.888	0.795	0.795
MLP	0.232	0.864	0.722	0.658	0.948	0.488	0.488
GBDT	0.226	0.913	0.835	0.785	0.929	0.736	0.736
ANN	0.522	0.906	0.857	0.891	0.821	0.895	0.895

ISSN: 1472-6947