Page 46 - AIH-1-3

Artificial Intelligence in Health Predicting mortality in COVID-19 using ML

Figure 8. Attribute importance ranking of the “DecisionTreeClassifier” method. Image created using Python’s Matplotlib library

A B

Figure 9. SHAP summary plots of the “DecisionTreeClassifier” method. (A) Barchart. (B) Beeswarm. Image created using Python’s Matplotlib library

TP 4.2. Model evaluation
TPR=
(TPFN)+ (IV) After running and evaluating all 324 different models, we
ranked them according to their scores. The ML models
FP achieved precision ranging from 90.09% to 93.76%, recall
FPR= from 83.42% to 96.99%, F1-score from 84.96% to 91.13%,
(FPTN)+ (V) AUC-ROC from 0.9003 to 0.9788, and runtime from 1.092
to 910.173 s. The results of this ranking are depicted in
Runtime is an additional metric that we used; it is Figure 19. The model with the highest score is positioned
measured in seconds and defined as the duration of a on the leftmost position, with the metric values decreasing
model’s iteration. as we move toward the right, so the last model scores the

Volume 1 Issue 3 (2024) 40 doi: 10.36922/aih.2591