Systematic Review of Automatic Assessment Systems for Resistance-training Movement Performance: A Data Science Perspective

The technical performance of resistance-training (RT) movement is commonly monitored through visual assessment and feedback by trained practitioners or by individual self-evaluation. However, both approaches are limited due to their subjectivity, inability to monitor multiple joints simultaneously, and dependency on the assessor's or exerciser's experience and skill. Portable data collection devices and machine learning (ML) have been combined to overcome these limitations by providing objective assessments for RT movement performance. This systematic review evaluates systems developed for providing objective, automatic assessment for RT movements used to improve physical performance and/or rehabilitation in otherwise healthy individuals. Databases searched included Scopus, PubMed and Engineering Village. From 363 papers initially identified, 13 met the inclusion and exclusion criteria. Information extracted from the collated papers included the experimental protocols, data processing, ML model development methodology and movement classification performance. Identified movement assessment systems ranged in classification performance (accuracy of 70%-90% for most classifiers). However, several methodological errors in the development of the ML models were identified, and additional aspects such as model interpretability or generalisability were often neglected. Future ML models should adopt the correct developmental methodology and provide interpretable and generalisable models for application in the RT environment.