Automated differentiation of ASD gait from normal gait patterns is important for early diagnosis as well as ensuring rapid quantitative clinical decision and appropriate treatment planning. This study explores the use of statistical feature selection approaches and artificial neural networks (ANN) for automated identification of gait deviations in children with ASD, on the basis of dominant gait features derived from the three#dimensional (3D) joint kinematic data. The gait data from 30 ASD children and 30 normal healthy children were measured using a state#of#the#art 3D motion analysis system during self#selected speed barefoot walking. Kinematic gait features from the sagittal, frontal and transverse joint angles waveforms at the pelvis, hip, knee, and ankle were extracted using time#series parameterization. Two statistical feature selection techniques, namely the between#group tests (independent samples t# test and Mann#Whitney U test) and the stepwise discriminant analysis (SWDA) were adopted as feature selector to select the meaningful gait features that were then used to train the ANN. The 10#fold cross# validation test results indicate that the selected gait features using SWDA technique are more reliable for ASD gait classification with 91.7% accuracy, 93.3% sensitivity, and 90.0% specificity. The findings of the current study demonstrate that kinematic gait features with the combination of SWDA feature selector and ANN classifier would serve as a potential tool for early diagnosis of gait deviations in children with ASD as well as provide support to clinicians and therapists for making objective, accurate, and rapid clinical decisions that lead to the appropriate targeted treatments. Artificial neural network, gait classification, gait feature, stepwise discriminant analysis. I. INTRODUCTION UTISM spectrum disorder (ASD) is the name refer to a group of neurodevelopmental disorders that affects the This work was supported by the Ministry of Higher Education (MOHE) Malaysia under the 2014 Federal Training (HLP) Scheme awarded to the main author. The research work was also funded by the MOHE through the Niche Research Grant Scheme (NRGS), project file: 600#RMI/NRGS 5/3 (8/2013) and (9/2013). C. Z. C. Hasan was with the Community College of Pasir Salak, Perak, Malaysia. She was on study leave from 2014 to 2017 at Faculty of Electrical Engineering, Universiti Teknologi MARA, 40450 Shah Alam, Selangor, Malaysia. She is now with the Department of Electrical Engineering, Politeknik Sultan Idris Shah, Sungai Lang, 45100 Sungai Air Tawar, Selangor, Malaysia (e#mail: zawiyah.hasan@gmail.com). R. Jailani and N. M. Tahir are with the Faculty of Electrical Engineering, Universiti Teknologi MARA, 40450 Shah Alam, Selangor, Malaysia (corresponding authors, phone: +603#55435052; fax: +603#55435077; e# mail: rozita@ieee.org and nooritawati@ieee.org). cognitive function and diminishes the quality of life of an individual. It is a severe and lifelong impairment, which can be identified in the early years of childhood. Recently ASD has become the most rapidly increasing neurodevelopmental disorder worldwide [1]. ASD is one of the most prevalent forms of developmental disabilities, with current estimates of prevalence is one in 68 children and it is 4.5 times more common among boys than among girls [2]. According to the fifth edition of the Diagnostic and Statistical Manual of Mental Disorders (DSM#5), children with ASD are recognized by persistent deficits in social communication and social interaction, in addition to the presence of restricted, repetitive behavior patterns, interests, or activities. Additional symptoms that support the diagnosis of ASD are the existence of movement and motor disturbances such as irregular motor signs, clumsiness as well as abnormal gait [3]. Earlier detection of these symptoms creates better opportunities for children with ASD to benefit more fully from early intervention or treatment programs [4]. It has been suggested that motor skills need to be considered and incorporated in early intervention programs [5]. Clinicians and researchers from various disciplines have identified movement and sensory disturbances as the focus symptoms of individuals with ASD [6]. Previous studies have reported a wide range of abnormal gait patterns in various aspects of gait parameters such as basic gait measurements, kinematic joint angles, and kinetic joint moments during walking in individuals with ASD [7]. Children with ASD were found to demonstrate a variety of significant alterations on the ankle and hip joint kinematics and kinetics [8]. Our recent study has also reported significant gait deviations in the 3D ground reaction forces of children with ASD which particularly related to the difficulties in supporting their body weight as well as exhibiting gait instability during the stance phase of gait [9]. Several significant ground reaction forces gait features can potentially be used in the identification of ASD gait [10]. An early identification of these aspects of gait deviations in ASD children is crucial in order to facilitate appropriate treatments and rehabilitation programs for the ASD patients requiring therapies. Nowadays, gait analysis is routinely used in clinical settings for the systematic study of the human walking patterns and also for the assessment of walking performance [11]. Gait can Use of statistical approaches and artificial neural networks to identify gait deviations in children with autism spectrum disorder Che Zawiyah Che Hasan, Rozita Jailani, and Nooritawati Md Tahir A INTERNATIONAL JOURNAL OF BIOLOGY AND BIOMEDICAL ENGINEERING Volume 11, 2017 ISSN: 1998-4510 74