The Application of Machine Learning in Activity Recognition with Healthy Older People Using a Batteryless Wearable Sensor
Zitong Zhou · 2020
The development of human activity monitoring has allowed for simulation and prediction in a variety of application scenarios. The battery-free wearable sensor is performed on sternum level clothing, mainly by script execution. It is essential to monitor human health, sleep, etc. With the increasing population in the world, the aging population is also becoming larger and larger. Hospitals and nursing homes are in great need to prevent injuries such as older people falling indoors. Existing methods construct the wireless wearable human body multi-physiological parameter monitoring system with wireless sensor network and central monitoring module unit. However, this system has a high false alarm rate and low accuracy in predicting human activities.To solve this problem, we constructed an activity detection method based on several machine learning algorithms, including Decision Tree,Gradient Boosting Decision Tree (GBDT), Adaboosting, Bagging and Random Forest. Although the decision tree is quite intuitionistic, ensemble learning models can improve the accuracy of prediction significantly by combining a lot of decision trees. By comparing the errors of each model, the most suitable monitoring model for predicting the physiological events of the elderly was selected. It is found that Random Forest outperforms the other methods, and achieves 99.97% on the training data and 98.8% accuracy on testing data, respectively.