Optimization of Sitting Posture Classification Based on User Identification Bruno Ribeiro*, Hugo Pereira, Rui Almeida, Adelaide Ferreira Departamento de Física, Faculdade de Ciências e Tecnologias, Universidade Nova de Lisboa, Quinta da Torre, 2829-516, Caparica, Portugal. bmf.ribeiro@campus.fct.unl.pt; p110595@campus.fct.unl.pt; rui.almeida@ngns-is.com; ajesus@fct.unl.pt Leonardo Martins* CA3, UNINOVA, Institute for the Development of New Technologies, Quinta da Torre, 2829-516, Caparica, Portugal, l.martins@campus.fct.unl.pt *Equal Contribution Claudia Quaresma, Pedro Vieira LIBPhys-UNL, Departamento de Física, Faculdade de Ciências e Tecnologias, Universidade Nova de Lisboa, 2829-516 Caparica, Portugal, q.claudia@fct.unl.pt; pmv@fct.unl.pt In a precursory work, an intelligent sensing chair prototype was developed to classify 12 standardized sitting postures using 8 pneumatic bladders (4 in the chair’s seat and 4 in the backrest) connected to piezoelectric sensors to measure inner pressure. A Classification of around 80% was obtained using Neural Networks. This work aims to demonstrate how algorithmic optimization can be applied to a newly developed prototype to improve posture classification performance. The aforementioned optimization is based on the split of users by sex and use two different previously trained Neural Networks (one for Male and the other for Female). Results showed that the best neural network parameters had an overall classification 89.0% (from the 92.1% for Female Classification and 85.8% for Male, which translates into an overall optimization of around 8%). Automatic separation of these sets was achieved with Decision Trees with an overall classification optimization of 87.1%. Sensory Intelligent Chair; Sitting Posture Classification; Machine Learning I. INTRODUCTION Various research groups have developed intelligent chairs by applying sheets of surface-mounted pressure sensors placed at the backrest and seat-pad of the chair using a 2D array arrangement of sensors or using techniques to find the best way to place unique force-sensitive resistors or even conductive textiles in the so called intelligent chair. These intelligent chairs have shown to be capable of detecting the presence of a person, detect their sitting posture and alert the user to improve their sitting posture or leave the chair due to a long sitting behaviour and can even be used as an input device to the office environment or even as a health monitoring device [1]–[10]. One of the biggest problems of the present society is that we spend long periods of time in a sitting position due to recent radical changes in the workplace, transportation, and even leisure activities [11]–[13]. One of the objectives of the aforementioned intelligent chairs is to alert the users of prolonged sitting behaviours and guide the users into positioning themselves in ‘correct’ sitting postures, because when a person is positioned in a sitting position, the ischial tuberosities, the thigh and the gluteal muscles do all the bodyweight supporting work, while the remainder of the weight is transferred to the ground by the feet, to the backrest and armrests when they are present [14]. But if a person has prolonged sitting behaviours, this situation will lead to a decrease of the lumbar lordosis [15], [16] which can then increase the physical risk factors related to of back, neck and shoulder pain [17], [18] due to anatomical changes and degeneration of intervertebral disks and joints [19], [20]. Bad posture while sitting can have a magnification effect over these health problems [21], which have been identified as one of the leading causes of work-related disability and loss of productivity in modern society [22], [23]. Clinically, there are several views of 'correct' and 'incorrect' postures, but only recently research groups were able to make quantitative studies to target this information. Some of those groups used multiple camera sensors to build a three- dimensional optical motion model of the subject and tried to determine the level of clinical advantage that the so-called 'correct' postures would provide over ‘incorrect posture’ [24], while other groups used adhesive 3D motion sensors applied on the skin to assess different spinal angles [25] to identify a ‘correct’ posture. Even recent studies have shown that there still exists a significant disagreement from rehabilitation 4 th Portuguese Meeting in Bioengineering, February 2015 Portuguese chapter of IEEE EMBS Faculty of Engineering of the University of Porto