Comparing accelerometer processing metrics for physical activity classification accuracy using machine learning methods

Musa, Sumayyah Bamidele (2022) Comparing accelerometer processing metrics for physical activity classification accuracy using machine learning methods. Masters thesis, Memorial University of Newfoundland.

[English] PDF - Accepted Version Available under License - The author retains copyright ownership and moral rights in this thesis. Neither the thesis nor substantial extracts from it may be printed or otherwise reproduced without the author's permission. Download (1MB)

Abstract

The purpose of this study was to compare the performance of three accelerometer processing metrics, Euclidean Norm Minus One (ENMO), ActiGraph Counts, and Monitor Independent Movement Summary (MIMS) units, in classifying physical activity using Random Forest (RF) and k-Nearest Neighbors (KNN) machine learning models, as well as to investigate the effect of hyperparameter tuning and feature selection on each processing metric. The dataset was sourced from a laboratory-based protocol involving raw acceleration data from 50 participants who held a smartphone device in their right hand while completing six activities. Findings indicated that even though the acceleration metrics performed well above 80% accuracy with both RF and KNN, the best performance was achieved with ENMO and the raw data as features. Additional accuracy of between 1% to 5% was achieved when the model hyperparameters were tuned before classification, and there was no difference when other features were included in the classification. In conclusion, ENMO is the best acceleration metric for classifying PA from accelerometers. Tuning the models and using a few selected features affected the models' accuracy.

Actions (login required)

View Item