Exploration of Digital Health Technologies

Model	Parameters	Accuracy	Sensitivity	Specificity	PPV	NPV
LSTM	Input = MFCC vector, loss = mean absolute error, optimizer = adam, activation = sigmoid	0.75 ± 0.06	0.29 ± 0.03	0.93 ± 0.02	0.61 ± 0.07	0.79 ± 0.01
CNN	Input = Mel-spectrogram images, input shape = (150,150,3), loss = binary crossentropy, optimizer = adam, activation = softmax	0.71 ± 0.03	0.44 ± 0.04	0.84 ± 0.03	0.49 ± 0.05	0.81 ± 0.01
HuBERT	Input = Encoder features	0.82 ± 0.03	0.50 ± 0.09	0.93 ± 0.05	0.73 ± 0.10	0.84 ± 0.02

Model

Parameters

Accuracy

Sensitivity

Specificity

PPV

NPV

LSTM

Input = MFCC vector, loss = mean absolute error, optimizer = adam, activation = sigmoid

0.75 ± 0.06

0.29 ± 0.03

0.93 ± 0.02

0.61 ± 0.07

0.79 ± 0.01

CNN

Input = Mel-spectrogram images, input shape = (150,150,3), loss = binary crossentropy, optimizer = adam, activation = softmax

0.71 ± 0.03

0.44 ± 0.04

0.84 ± 0.03

0.49 ± 0.05

0.81 ± 0.01

HuBERT

Input = Encoder features

0.82 ± 0.03

0.50 ± 0.09

0.93 ± 0.05

0.73 ± 0.10

0.84 ± 0.02