Activity Recognition Using Wrist-Worn Sensors for Human Performance Evaluation

Recent development of wearable technology has opened up great opportunities for human performance evaluation applications in various domains. In order to measure the physical activities of an individual, wrist-worn sensors embedded in smartwatches, fitness bands, and clip-on devices can be used to collect various types of data, as the subject performs regular daily activities. In this paper, we propose using the acceleration data generated by wrist-worn sensors to recognize ambulation activities for performance evaluation purposes. Twelve features are extracted from the raw accelerometer data, then feed into individual classifiers as well as their combinations for training and validation. The classifiers we consider in this paper are Naive Bayes, Support Vector Machines, Decision Tree, k-Nearest Neighbors, Multilayer Perceptron, and Random Forest, as they have been reported effective by a few previous works. We evaluate the set of selected classifiers with real-world data sets generated by three subjects, including data from both left and right wrists. The best accuracy performance is the combination of classifiers which is above 90%. The result shows that data generated by wrist-worn accelerometer sensor is sufficient for ambulation activity recognition and can be used for human performance evaluation applications.