Electromyogram-based classification of hand and finger gestures using artificial neural networks

Ji Young Min (github: wnet500) and Kyung Hyun Lee (github: Jovinus) contributed equally to this work.

The codes for 2.6 Modeling in Materials and Methods of the paper are available.

*** Note that all codes are executable ONLY if your own data exist

Datasets

• Here, the datasets we used in the paper can not be released for personal information protection.

• Instead, you can identify a sample dataset. Please refer to sample_dataset.csv

  • sample_dataset.csv shows the examples of the datasets used for modeling (Note that this is not a real subject's dataset)
  • In the "label" column, 0: rest, 1: rock, 2: scissor, 3: paper, 4: one, 5: three, 6: four, 7: good, 8: okay, 9: gun

Modeling

1. ANN gridsearch

• This code covers the gridsearch process and training with the best params for ANN

• ANN gridsearch was conducted with TensorFlow 2.0

• See ANN_gridsearch.py

• You can see the sample tensorboard results of gridsearch with the mean of cv accuracies here.
  (this is the same result with tensorboard --logdir=./logs/hparam_tuning_results if you run the code with your own datasets)

• You can see the sample tensorboard results of gridsearch with accuracy and loss changes according to epochs for each cv here.
  (this is the same result with tensorboard --logdir=./logs/hparam_tuning if you run the code with your own datasets)

2. SVM, RF, LR gridsearch

• This code covers the gridsearch process and training with the best params for SVM, RF, and LR

• ML gridsearch was conducted with scikit-learn ver. 0.23.2

• See ML_gridsearch.py