Detecting cognitive impairment in cerebrovascular disease using gait, dual tasks, and machine learning

Table 3 Performance on test sets using gait metrics from the accelerometer (in %)

Condition	SS	DAn	DS1	DS7	F	DAn + SS^a)	DS1 + SS^b)	DS7 + SS^c)	F + SS^b)
Classifier	LR	LR	SVM	LR	LR	RF	LR	LR	RF
Feature Sel.	None	RF	None	None	None	RFE	RFE	KBest	KBest
Sampling	None	SMOTE	SMOTE	None	None	None	None	SMOTE	None
# Features	3.0 ± 0.0	2.0±0.0	3.0±0.0	3.0±0.0	3.0±0.0	2.0±0.0	4.6±1.0	4.5±1.0	3.4±1.7
Bal. Acc.	69.5	64.5	62.6	61.2	60.4	66.9	70.6	68.1	66.1
Sensitivity	72.4	64.3	58.6	51.7	62.1	75.0	69.0	65.5	79.3
Specificity	66.7	64.7	66.7	70.6	58.8	58.8	72.2	70.6	52.9
F1 score	75.0	69.2	65.4	61.2	66.7	75.0	74.1	71.7	76.7
AUC	66.3	65.5	65.7	57.4	63.1	63.4	75.5	66.5	57.8

SS: Single-task; DAn: Walking while naming animals; DS1: Serial subtraction by 1s; DS7: Serial subtraction by 7s; F: Walking Fast; LR: Logistic Regression; RF: Random Forest; SVM: Support Vector Machine; RFE: Recursive feature elimination; AUC: Area under the curve. ^a) using dual-task costs; ^b) using dual-task costs (or capacity indexes) and SS metrics; ^c) using dual-task costs and dual-task metrics

ISSN: 1472-6947