What algorithmic approach underlies activity classification in most fitness trackers?

The algorithmic approach that underlines activity classification in most fitness trackers is based on convolutional neural networks (CNNs), particularly utilizing accelerometer data. According to one study, recent advancements emphasize using machine learning techniques to analyze data collected from wearable inertial measurement units, which can reliably classify physical activities^[2].

Specifically, real-time human activity recognition leveraging CNNs from accelerometer data is a prevalent method, as reported in various research findings^[5]. This reveals a trend where deep learning models, like CNNs, are effectively employed to enhance the accuracy of activity classification.