Unveiling the sound of the cognitive status: Machine Learning-based speech analysis in the Alzheimer’s disease spectrum

Table 6 Hyperparameters of classification and regression models

Model	Parameter^a	HPO^b	Value/search space^c
Random forest	Number of estimators	-	200
	Class weight^d	-	Balanced
	Max depth	TPE	{2, \(\ldots\), 10}
	Min samples split	TPE	{2, \(\ldots\), 40}
	Min samples leaft	TPE	{2, \(\ldots\), 30}
	Max samples	TPE	[0.5, 1.0]
	Max features	TPE	[0.5, 1.0]
XGBoost	Number of estimators	-	200
	Max depth	TPE	{2, \(\ldots\), 10}
	Learning rate	TPE	[0.01, 0.3]
	Gamma	TPE	[0.0, 100.0]
	Min child weight	TPE	[0.0, 100.0]
	Subsample	TPE	[0.2, 1.0]
	colsample_bytree	TPE	[0.2, 1.0]
	colsample_bynode	TPE	[0.2, 1.0]
	L1 regularization	TPE	[0.1, 10.0]
	L2 regularization	TPE	[0.1, 10.0]
	Scale positive weight^d	TPE	[0.1, 10.0]
Support vector machines	Kernel	-	Polynomial
	C	TPE	[1e-05, 1e02]
	Degree	TPE	{1, \(\ldots\), 10}
	Class weight^d	TPE	[0.05, 0.95]
	Coef\(_0\)	TPE	[0.0, 10.0]
K-nearest neighbors	Number of neighbors	Grid search	{4, \(\ldots\), 30}
K-nearest neighbors	Weights	Grid search	{Uniform, Distance}

For the hyperparameter optimization conducted using tree-structured Parzen estimator (TPE), a total of 1000 configurations were sampled, with the initial 500 being randomly selected
^aHyperparameters not listed in this table were selected at their default value
^bHyperparameters that were left fixed are specified by “-”
^c\(\{\cdot \}\) indicates sampling of categorical values, while \([\cdot ]\) indicates sampling of real values
^dParameter only considered for classification problems

ISSN: 1758-9193