Modern Machine-Learning Predictive Models for Diagnosing Infectious Diseases

Table 5

Comparison of the reviewed articles concerning the applied ML techniques and the resulted performance.

Reference	ML technique	Model performance
[19]	SVM (Gaussian kernels)	For HFMD dataset (AC: 70.9%, precision: 60.6%, SP: 78.0%, F1 score: 55.7%, and recall: 55.9%); For tetanus dataset (AC: 80.2%, precision: 78.4%, SP: 53.4%, F1 score: 86.0%, and recall: 98.1%)
[20]	NB, filtered classifier (FC), and RF	Success ratio with the weight of each medical variable (vital signs) that affects the prediction
[21]	LR	AUC: 0.68, recall: 60%, specificity: 75%, precision: 25%, NPV: 93%, F1 score: 35%, and MCC: 25%
[22]	SVM	PFA: 30%, F1 score: 74%, precision: 71%, and recall: 78%
[23]	Unsupervised sentiment analysis	Precision: 77.3%, recall: 68%, F1 score: 72.4%
[24]	NB	ROC: 89.91%, SN: 47%, and SP: 37%
[25]	XGBoost algorithm	AUC: 0.97, SN: 81.9%, SP: 97.9%
[26]	SVM	AC: 91.18%, SN: 100%, SP: 84.21%, PPV: 83.33%, F1 score: 90.91%, and AUC: 0.958
[27]	RF	AUC: 0.847, SN: 84.24%, SP: 91%, HLT statistics: 12.779, and HLT : 0.120
[28]	EHR-based model (the study did not mention the used algorithm)	SN: 41.7%, SP: 96.7%, and PPV: 41.7%
[29]	NB and LR	AUC: 0.93, BSS: all classifiers achieve positive BSS scores
[30]	XGBoost algorithm	AUC: 0.95, AC: 88%, SN: 88%, and SP: 89%
[31]	Ensemble learning model	AC: 99.73%, precision: 99.46%, recall: 100%, F1 score: 99.73%, and AUC: 0.9973
[32]	NB	AUC greater than: 98%, SN: 44.44% for hepatitis E and 96.67% for measles, SP: 96.36% for dengue fever and 100% for 5 diseases, median of total accuracy: 97.41%, and -index:0.960

SN: ; AUC: area under the curve; SP: specificity; PPV: positive predictive value; AC: accuracy; PFA: probability of false alarm: BSS: brier skill score; HLT : the value of the Hosmer–Lemeshow test; ROC: receiver operating characteristic.