Experiment 1.

7 downloads 162623 Views 1MB Size Report
postersession.com. • Augmentative and Alternative Communication (AAC) apps enable non- ... development of AAC applications to assist individuals with ASD,.
An exploratory analysis targeting diagnostic classification of AAC app usage patterns Adham Atyabi, Beibin Li, Yeojin Amy Ahn, Minah Kim, Erin Barney, Frederick Shic Seattle Children’s Innovative Technologies Lab (SCITL), Seattle Children’s Research Institutea Department of Pediatrics, University of Washington Introduction

FreeSpeech

• Augmentative and Alternative Communication (AAC) apps enable nonspeech communicative forms. • Speech-generating devices (SGDs) -icons/pictures are tapped to produce spoken words. • SGD apps are widely used to support communication and language learning for individuals with disabilities such as autism spectrum disorder (ASD. • Objective: To evaluate several feature representations of usage patterns, coupled with data mining and data modelling techniques, for identifying differences in AAC usage patterns between users with and without ASD. • The study was conducted using data streams aggregated from an AAC app called FreeSpeech.

Problem Statement

Experimental Design & Results

• FreeSpeech was an AAC iPad application employed to foster communication skills and help people with speech and language disorders to communicate with others. • Nonverbal communication was facilitated through the selection of a sequence of words/images that shape meaningful phrases. Figure 1 depicts a screenshot of this app in action. • The application was released in January 2012 and terminated in May 2014. Overall 6033 individuals used the application. • Current analysis is performed using the data from 5372 users (Jan. 2012-Jan. 2014). • Given that the app’s usage information was originally collected for developer’s debagging purposes and the application was not meant to be employed as a diagnosis or treatment tool, little attention been given to the data structure and representation during the app development. • Diagnosis information was collected via self-report survey in the app (see a graphical representation in Figure 2).

• Although substantial effort and attention is diverted towards development of AAC applications to assist individuals with ASD, these apps are not designed with usage analytics in mind. • The development of AAC applications mostly focuses on providing an accessible mechanism for individuals to have daily and routine trainings to improve their performances. • Analysis of users’ performance is usually done as a self-assessment or is conducted by caregivers and medical professionals using psychological/physical/clinical measures. • This study is focuses on • exploring the existing data mining and machine learning data analysis procedures, • identifying suitable mechanisms for data representation of such usage recordings, • assessing the existing potential of such procedures for prediction of users’ medical diagnoses.

Fig 1: A screenshot from the FreeSpeech app

Fig. 2: Distribution of FreeSpeech app users based on their self-disclosed diagnoses

• As reported in Table 1, the initial statistical analysis of usage patterns only found a single, weakly significant difference between the ASD and non-ASD groups in the frequency of word selection (p