Table 5.

Classifier hyperparameters, optimised using leave-p-out nested cross-validation.

Hyperparameters	Classifier	Range
Regularisation Strength (α1)	LR, SVM	10i where, i = − 7, −6, …, 6, 7
l1 penalty (α2)	LR	0 to 1 in steps of 0.05
l2 penalty (α3)	LR, MLP	0 to 1 in steps of 0.05
Kernel Coefficient (α4)	SVM	10i where, i = − 7, −6, …, 6, 7
No. of neighbours (α5)	KNN	10 to 100 in steps of 10
Leaf size (α6)	KNN	5 to 30 in steps of 5
No. of neurons (α7)	MLP	10 to 100 in steps of 10
No. of convolutional filters (β1)	CNN	3 × 2k where k = 3, 4, 5
Kernel size (β2)	CNN	2 and 3
Dropout rate (β3)	CNN, LSTM	0.1 to 0.5 in steps of 0.2
Dense layer size (β4)	CNN, LSTM	2k where k = 4, 5
LSTM units (β5)	LSTM	2k where k = 6, 7, 8
Learning rate (β6)	LSTM, MLP	10k where, k = − 2, −3, −4
Batch Size (β7)	CNN, LSTM	2k where k = 6, 7, 8
Epochs (β8)	CNN, LSTM	10 to 250 in steps of 20