Machine Learning for Training and Physiotherapy

Not too long ago, wearable devices to track the physical activity of individuals such as fitness trackers, running watches, etc., became popular and conquered the market.
Obviously, such devices are a help for many people to motivate themselves to be physical active. Moreover, it is a way to evaluate the physical fitness and to track personal progress.
By means of kinematic recordings of different exercises for upper limb strengthening, this thesis shows how the basic idea of tracking physical activity can be extended to the area of physiotherapy.

Therefore, three classification procedures are elaborated in order to
approach the problem from different angles. Classical machine learning
algorithms (Logistic Regression, Linear Discriminant Analysis, Random Forest
Classifier, Support Vector Classifier) as well as Neural Networks are taken into account. Individual strengths
and weaknesses of the different approaches are identified.
It is shown that automatic detection and classification of physiotherapeutic exercises is possible with high accuracy. Even single exercise repetitions can be detected with two of the three elaborated approaches.