EMG analysis and modelling of flat bench press using artificial neural networks

Abstract

The objective of this study was to evaluate the contribution of particular muscle groups during the Flat Bench  Press (FBP) with different external loads. Additionally, the authors attempted to determine whether regression  models or Artificial Neural Networks (ANNs) can predict FBP results more precisely and whether they can optimise the training process. A total of 61 strength-trained athletes performed four single repetitions with 70, 80, 90 and 100% of one repetition maximum (1RM). Based on both kinematic and electromyography results, a regression model and ANNs for predicting the FBP performance was created. In an additional study, 15 athletes performed the training session in order to verify the created model. The results of the investigation show that the created neural models 9-4-1 structure (NRMSE [Normalised Root Mean Squared Error], for the learning series was 0.114, and for the validation and test series 0.133 and 0.118, respectively), offer a much higher quality of prediction than a non-linear regression model (Absolute regression error – Absolute network error =47kg–17kg=30kg).

Key words: Non-linear models; Artificial neural networks; Bench press performance; Electromyography.