EURASIP Journal on Audio, Speech, and Music Processing

Table 10 Accuracy, precision, recall, and F1 scores of the classifiers with i-vector features are shown using MIM, mRMR, and JMI feature selection algorithms. For each feature selection algorithm, results for only the best performing feature sizes are shown. Experiments were done using the conversation-based approach. Systems with best F1 score are shown in bold

From: Automatic detection of attachment style in married couples through conversation analysis

	I-vector features
	Feature type	Accuracy	Precision	Recall	F1 score
SVM	MIM (30)	0.64	0.66	0.57	0.61
	mRMR (10)	0.62	0.62	0.58	0.6
	JMI (15)	0.72	0.75	0.65	0.7
Decision tree	MIM (15)	0.76	0.79	0.7	0.74
	mRMR (5)	0.74	0.75	0.72	0.74
	JMI (50)	0.65	0.65	0.62	0.64
Random forest	MIM (30)	0.76	0.75	0.76	0.76
	mRMR (15)	0.74	0.74	0.72	0.73
	JMI (50)	0.77	0.77	0.76	0.76
AdaBoost	MIM (15)	0.77	0.77	0.76	0.76
	mRMR (30)	0.7	0.69	0.72	0.7
	JMI (15)	0.79	0.77	0.81	0.79
Gradient boosting	MIM (15)	0.79	0.77	0.81	0.79
	mRMR (15)	0.73	0.72	0.72	0.72
	JMI (15)	0.81	0.82	0.79	0.8
Extra tree	MIM (30)	0.79	0.77	0.81	0.79
	mRMR (15)	0.76	0.75	0.77	0.76
	JMI (30)	0.8	0.8	0.77	0.79
XGBoost	MIM (10)	0.73	0.72	0.74	0.73
	mRMR (10)	0.73	0.73	0.7	0.72
	JMI (15)	0.76	0.78	0.72	0.75

Back to article page