Identification of Activities of Daily Living through Artificial Intelligence: an accelerometry-based approach

The accelerometer is available on most of these mobile devices. It allows the acquisition and calculation of different physical parameters. Due to the use of pattern recognition, it also enables the identification of several Activities of Daily Living (ADL), such as walking, running, going downstairs, going upstairs, and standing. The feature extraction step performs the extraction of the five most significant distances between peaks, the average, standard deviation, variance and median of extracted peaks and raw data, and the maximum and minimum of raw data. The focus of this paper is the implementation of multiple artificial intelligence methods for the recognition of ADL, including Logistic Regression, Combined nomenclature rule inducer, Neural Network, Naive Bayes, Support Vector Machine, Decision Tree, Stochastic Gradient Descent, and k-Nearest Neighbor. The Decision tree reported the average accuracy of 85.22% between classes. This method also presents an F1-score value of 85,13% and a precision value of 85,08%. Nevertheless, the study has limitations associated with the use of mobile devices. The position and location of the device in the data collection phase need further investigation, and the system architecture demands higher energy consumption.