Table 3.

Results of the AutoML experiments for cognitive test performance classification with different feature subsets used for training.

Feature subset	Accuracy	F1 score	Log loss	Precision	Recall	AUC (PR)	AUC (ROC)
Stroke-176	65.5%	0.735	0.630	68.3%	79.6%	0.581	0.671
Stroke-165	65.8%	0.486	0.631	60.5%	40.6%	0.582	0.668
Stroke-11	64.2%	0.395	0.646	60.4%	29.3%	0.527	0.641

Data split: 80% training, 10% validation and 10% testing. Only stroke-based features are considered, because the AutoML approach requires at least 1,000 rows of data.