Table 7.

Frequency of different machine learning (ML) techniques across the 92 reviewed papers. Convolutional neural network (CNN)–based architectures were the most common. The total count exceeds 92 because many studies used multiple techniques (eg, a CNN for feature extraction and a long short-term memory (LSTM) for temporal analysis).

Frequency of machine learning techniques used	Studies (n=92), n
Unsupervised & self-supervised learning	7
Time series & probabilistic methods (DTWa, HMMb, etc)	9
Transformers & attention mechanisms	11
RNNc-based architectures (LSTM, etc)	14
Object detection and segmentation (YOLOd, U-Nete, etc)	16
Classical ML (SVMf, random forest, KNNg, etc)	29
CNN-based architecture (ResNeth, I3Di, TCNj, etc)	55

^aDTW: dynamic time warping.

^bHMM: hidden Markov model.

^cRNN: recurrent neural network.

^dYOLO: You Only Look Once.

^eU-Net: U-shaped convolutional network.

^fSVM: support vector machine.

^gKNN: k-nearest neighbors.

^hResNet: residual network.

ⁱI3D: inflated 3D convolutional network.

^jTCN: temporal convolutional network.