Long short-term memory training for the assessment of vigilance
The assessment of vigilance is of increasing importance within our 24/7 working society. Posturography is one candidate for a quick, mobile and cost-efficient vigilance assessment. Nevertheless classification accuracy is yet insufficient. This contribution aims at improving classification accuracy of posturographical vigilance assessment by utilizing the information hidden within temporal dynamics of feature sequences. For this purpose a Recurrent Neural Network, Long Short-Term Memory (LSTM), is applied. In order to evaluate whether temporal dynamics offer additional information, results from LSTM training are compared to non-recurrent approaches. Results indicate that there is no significant gain in accuracy achieved by learning temporal dynamics.