Internet of things f or optimerad och smart prestationsutveckling

Abstract

The growing market for digital aids such as smart watches and heart rate monitors shows an interest in individual training optimization. This report aimed to develop a basic prototype of a mobile application with a focus on intuitively presenting analyzed training data to the end user and a study comparing performance between a mathematical model and machine learning models. The project was limited to using existing hardware and instead focused on data analysis. The study was limited to not taking into account the subjects' nutritional intake or sleep, with the exception of the number of hours of sleep. The application was structured as a client server solution where the server side was developed as a REST API in Python with the framework Flask and the application side was developed using react native. The study was conducted over 40 days with 6 subjects with di erent training backgrounds aged 23.2 1.6 years. All were to report values for rSPE every day and were required to exercise at least 4 times a week with varying intensity. The data from the study were processed and then used in three models: the Banister model and two machine learning models, an ANN model and an RNN model. MSPE was used to measure the models' ability to predict HRV. The mean value of the MSPE standard deviation for the Banister model was 0:066 0:052, for ANN and RNN it was 0:127 0:141 and 0:05 0:005, respectively. RNN demonstrated a su ciently low MSPE value without over-adapting the data and was therefore designated as a suitable alternative to the Banister model, which received the lowest MSPE but instead over-adapted the data. The application was mainly completed but lacks some desired functionality. Functionality that was implemented in both the mobile application, the server and the database was: user management with login, input of sRPE and HRV as well as creation of training sessions and demonstration of predicted HRV. The application works as a prototype that can be further developed in combination with further studies with longer time intervals and more subjects to further investigate the possibilities of using machine learning for training optimization.