Journal of Medical and Biological Engineering, 33(4): 406-414 406 Three-layer Activity Recognition Combining Domain Knowledge and Meta-classification Simon Kozina * Hristijan Gjoreski Matjaž Gams Mitja Luštrek Department of Intelligent Systems, Jožef Stefan Institute, Ljubljana SI-1000, Slovenia Received 24 Sep 2012; Accepted 18 Jan 2013; doi: 10.5405/jmbe.1321 Abstract One of the essential tasks of healthcare and smart-living systems is to recognize the current activity of a particular user. Such activity recognition (AR) is demanding when only limited sensors are used, such as accelerometers. Given a small number of accelerometers, intelligent AR systems often use simple architectures, either general or specific for their AR. In this paper, a system for AR named TriLAR is presented. TriLAR has an AR-specific architecture consisting of three layers: (i) a bottom layer, where an arbitrary number of AR methods can be used to recognize the current activity; (ii) a middle layer, where the predictions from the bottom-layer methods are inputs for a hierarchical structure that combines domain knowledge and meta-classification; and (iii) a top layer, where a hidden Markov model is used to correct spurious transitions between the recognized activities from the middle layer. The middle layer has a hierarchical, three-level structure. First, a meta-classifier is used to make the initial separation between the most distinct activities. Second, domain knowledge in the form of rules is used to differentiate between the remaining activities, recognizing those of interest (i.e., static activities). Third, another meta-classifier deals with the remaining activities. In this way, each activity is recognized by the method best suited to it, leaving unrecognized activities to the next method. This architecture was tested on a dataset recorded using ten volunteers who acted out a complex, real-life scenario while wearing accelerometers placed on the chest, thigh, and ankle. The results show that TriLAR successfully recognized elementary activities using one or two sensors and significantly outperformed three standard, single-layer methods with all sensor placements. Keywords: Activity recognition, Ambient intelligence, Intelligent healthcare, Machine learning, Meta-classification, Multi-layer activity recognition 1. Introduction The world’s population is aging rapidly, threatening to overwhelm society’s capacity to take care of its elderly members. The percentage of persons aged 65 or over in developed countries is projected to rise from 7.5% in 2009 to 16% in 2050 [1]. This is driving the development of innovative healthcare and smart-living technologies to help the elderly live independently for longer and with minimal support from the working-age population [2,3]. To be used in a real-world setting, healthcare and smart-living systems must take into account the user’s situation and context, making activity recognition (AR) an essential component of such systems [4,5]. AR requires a sensor system that observes the user and intelligent software that infers the user’s activities from the sensor data [6,7]. The idea for the AR method proposed here was initiated * Corresponding author: Simon Kozina Tel: +386-1-4773195; Fax: +386-1-4773131 E-mail: simon.kozina@ijs.si and gradually developed in two European healthcare projects: Confidence [8] and CHIRON [9]. Even though AR was not the main goal in either of the projects, it eventually emerged as one of the most important components, being the foundation for further reasoning in the main tasks, including the detection of falls, the detection of unusual behavior, and the estimation of human energy expenditure [10-12]. The initial AR model developed in the Confidence project was a traditional, single-layer classification model that uses two types of sensor (accelerometers and location sensors) to achieve adequate performance [13]. In the CHIRON project, the AR model was upgraded to two layers, which improved performance. The present study presents a three-layer architecture for AR called TriLAR. The TriLAR architecture consists of the following: a bottom layer (an arbitrary number of independent AR methods), a middle layer (hierarchical structure that aggregates the predictions from the bottom-layer methods), and a top layer (a hidden Markov model (HMM) that uses the temporal dependence of activities to remove spurious transitions between them). This architecture was tested on a dataset recorded using ten volunteers who acted out a complex, 90- minute scenario while wearing accelerometers placed on the