A Comparative Analysis of Machine Learning Algorithms for Accurate Step Detection in Wrist Worn Devices

This paper presents a comprehensive study on the application and evaluation of machine learning algorithms for accurate step detection on wearables, with a particular focus on smart wristbands. A unique data collection system was developed using an inertial sensor and an ESP-8266 kit to acquire both raw and processed accelerometer and gyroscope data. The collected data was transmitted via the MQTT protocol and then processed and labeled to indicate whether the step was taken or not. Five different machine learning algorithms such as AdaBoost, decision tree, k nearest neighbors, Naive Bayes, and random forest were used for step detection. In addition, 5-fold cross validation was used to ensure the reliability of the machine learning algorithms. The performance of each model was evaluated using the metrics accuracy, precision, recall, F1-score, receiver operating characteristic (ROC) curve, area under the ROC curve. Comparative analysis revealed that machine learning algorithms can significantly improve step detection accuracy in wearable devices. The findings contribute to the advancement of health monitoring, fitness tracking and rehabilitation applications. Future studies aim to collect and analyze data with more sensor variations across a broader demographic.