Sensors and Machine Learning Algorithms for Location and POSTURE Activity Recognition in Smart Environments

Human activity recognition (HAR) has become a focus of study over the past few years. It is widely used in many fields like health, home safety, security, and energy saving, among others. Research around the health area has evidenced an important increase and a promissory impact on the life quality of a population like the elderly. If we combine sensors and a health condition then we may have a technological solution with methods and techniques that will help us to improve life quality. Smart sensors have become popular. They allow us to monitor data and acquire data in real-time. In HAR, they are used to detect actions and activities like breathing, falling, standing up, or walking. Many commercial solutions use this technology in real-life applications. However, we focused this paper on the Vayaar sensor and the WideFind sensor, two commercial sensors based on ultra-wideband technology, with promising performance, as part of a study developed at the Human Health and Activity Laboratory (H2AL) in the Luleå Tekniska Universitet in Sweden. The study performed a technological and commercial comparison applying machine learning techniques in WEKA for two datasets created with the data gathered from each sensor during an experiment, in which precision and accuracy were analyzed as evaluation parameters of the applied methods. It was identified that random forest (RF) and LogitBoost were the most suitable classifiers to process both WideFind and Vayyar datasets. Random forest had a performance of 85.99% of precision, 85.48% of recall, and 96% of ROC area for the WideFind sensor while LogitBoost had a 69.39% of the performance for precision, 68.89% for recall, and 88.35% of ROC area for the Vayaar sensor.