This Jupyter notebook demonstrates the use of machine learning to detect the intensity of physical activity using the readings of wearable devices that can track body motions and basic physiological parameters such as heart rate.
Many people have trouble motivating themselves to do vigorous exercise, either because they believe they are already active
enough or, worse yet, they don’t know how much is enough. A smart device that can detect the level of the physical activity of an individual over time would help to keep track the stretches of intense physical activity performed. Moreover, such an exercise tracking application would allow a user to monitor the progress towards a set goal for the amount of exercise to perform in a given period of time.
The preliminary analysis I am presenting here shows a direct correlation between the heart rate and acceleration measured by the device with intensity levels of physical activity. This model could become the engine of an app to monitor and record over time the level of physical activity of a person using a set of readings from a smartphone and/or other wearable devices that that person uses in his or her daily life.
In this study I am using the PAMAP2 dataset from the UC Irvine Machine Learning repository to build the model. The dataset if accessible through this URL:
https://archive.ics.uci.edu/ml/datasets/PAMAP2+Physical+Activity+Monitoring
The dataset includes readings of a total of 52 body vitals monitored on several parts of the body of the subjects that were studied along with sensory data such as accelerometer, magnetometer, and gyroscope data that precisely describe the movement in which the subjects are involved when the readings are taken. The readings are associated to particular type of activity in which the subjects are involved. I chose to group these activities into four groups labeled minimum, low, medium, and high intensity activities. The goal of the data analysis project is to apply a machine learning techniques to develop a model that correlates sensor readings to the activity level. The final objective is to deploy the model as an app that tracks the intensity of body activity over time using readings of vital signs and body motion from personal smart devices.
In this preliminary study I am using a support vector machine classifier to learn the intensity of physical activity starting from the training data. The initial tests show a promising accuracy of at least 80 %. As next steps I plan to test other important descriptors that might be relevant to obtain an optimal model. I will also test other classifiers such as neural networks.