Proceedings of International Joint Conference on Neural Networks, Orlando, Florida, USA, August 12-17, 2007 A hybrid Support Vector Machine and autoregressive model for detecting gait disorders in the elderly Daniel T.H. Lai, Ahsan Khandoker, Rezaul K. Begg and M.Palaniswami Abstract- The consequence of tripping and falling in the elderly population is serious because of the life threatening fractures which occur and the high medical costs incurred. Recently, the minimum toe clearance (MTC) has been employed in gait analysis as a sensitive gait variable for early detection of elderly people at risk of falling. In previous work, we successfully applied statistical and wavelet analysis methods with Support Vector Machines (SVM) to model the risk of tripping in the elderly. In this work, we propose to model the MTC time series as a wide based stationary random signal using the autoregressive (AR) process. Initially, it was found that a fourth order AR model constructed from 512 MTC samples per subject on 23 subjects completely modelled the balance impaired gait (pathological) from normal gait. However, when the number of MTC samples were reduced to 32, the two groups became inseparable. We then proposed a hybrid system consisting of a SVM classifier with AR model coefficients as input features to separate the two classes. It was found that SVMs with linear and Gaussian kernels produced 100% leave one out accuracies without the need for prior feature selection algorithms. In contrast, SVM models built previously from the best set of wavelet features produced only 86.95% leave one out accuracies. These results suggest that pathological gait is best modelled by the AR process if sufficient MTC data is available. In the case of shorter MTC data, the AR model still provides powerful and robust discriminative features which can be used by the SVM to detect elderly people at risk of falling. I. INTRODUCTION The ageing population of highly industrialized countries such as Australia is on the rise. It is estimated that by the year 2035, 22% of the Australian population will be aged over 65 years [1] compared to just 12% in 2005. This forecasted increase requires continuous improvements to the current health care system to cope with the pathologies or diseases that may affect the elderly population. One of the more common afflictions in this age group is falls. Falling is a serious health problem known to affect 53% [2] of the elderly population with tripping and slipping being the major etiologies i.e., 83% of elderly falls [3]. Trip related falls potentially result in life threatening fractures and are known to cost the Australian government approximately 2.4 billion AUD [4] annually in injury costs. Research in automated detection and diagnosis of those at risk of falling is therefore essential so that preventive or rehabilitative measures can be taken. Daniel T.H. Lai, Ahsan Khandoker and M.Palaniswami are with the Department of Electrical and Electronic Engineering, The University of Melbourne, Parkville Campus, Vic 3010, Australia (phone: +61-3-8344 4942; email: d.lai 0ee.unimelb.edu.au ). Rezaul K. Begg is with the Centre for Ageing, Rehabilitation, Exercise and Sport, Victoria University, Vic 8001, Australia. It has been proven that ageing changes gait patterns which destabilize the human locomotion balance mechanism. Falls due to tripping in the elderly are the result of declines in the balance control function due to ageing [5], [6]. Numerous studies undertaken in recent years focussed on identifying suitable gait variables that are affected as a result of the ageing process and also those that best indicate declines in gait performance. Contemporary gait variables using kinematics e.g., measurements of joint angles [7]-[10] and kinetics such as time-distance variables (e.g., walking speed, stance/swing times, step length) have long been used to identify key characteristics of gait degeneration in the elderly. Pavol et. al [11] have studied mechanisms leading to falls in the elderly and the natural recovery strategies employed to prevent falling. Barak et al. [12] investigated the variance of kinematic variables such as stride frequency, stride length, walking speed, lateral body sway etc. as likely causes of falling. Both these studies and several others have concluded that large variations in gait variables are risk indicators of falling, however it is still unknown as to whether they directly affect tripping or if tripping is actually the result of body adjustments to avoid future falls. Recently, a more sensitive gait variable known as mini- mum toe clearance (MTC) has been used to describe age- related declines in gait with better success as a predictor of falls risk [13], [14]. MTC during walking (see Fig. 1) is defined as the minimum vertical distance between the lowest point on the shoe and the ground during the mid-swing phase of the gait cycle. This variable is important because it determines successful negotiation of the environment in which we walk. The literature has also suggested a decrease in MTC height (1. 1 cm) with ageing [6], thereby providing a strong rationale for MTC being associated with tripping during walking. 5 9 13 17 21 25 29 33 37 41 45 49 53 57 61 65 69 73 77 Time (10ms) Fig. 1. Left foot (toe) clearance over the swing phase showing the MTC event in one gait cycle. In previous work, SVM models based on statistical [15] 1-4244-1 380-X/07/$25.00 ©2007 IEEE E E .0 a MTC Ground Level