Data Collection And Preliminary Recognition Of Autism Spectrum Disorder Early Signs In Children’s Toy Playing

Autism Spectrum Disorder(ASD)has a high morbidity all over the world(approximately1of 88 children according to US CDC,2012),may be characterized by social interaction disorder,verbal and non-verbal communication disorder and repetitive or stereotyped behaviors,which possibly lasts for the whole life.Its dangerous impact for kids worries their families.Because of its unknown etiology,occurrence cannot totally be avoided at present.However,when ASD is found earlier,there will be higher possibility by treatment to take the kids back to the normal life.To get the early signs of ASD,the development of sensor network today contributes one of effective ways which this article will introduce,is using sensors to collect data from the toys which infants play,then using computer to analyze.Usually the kids with ASD show more non-functional repeated play than others.The data will first be transmitted by BLE(Bluetooth Low Energy)to and stored in the mobile phone,to make the collection can easily to be taken place in the kindergarten or infants’ home while decreasing the expense of devices.In experiments,we set CC2541 Sensor Tag from Texas Instruments in the toys to collect acceleration or angular velocity data and developed a software on Android platform to collect and store the data.To obtain some useful information,this research is trying to do the data mining and classification based on machine learning.The way that infants play with toys shows very high randomness and uncertainty.Under the non-external disturbance experiment conditions,infants hardly repeat their actions.Defining the activities which infants are doing,the time this research should divide these different activities and increasing the number of samples of activities are big challenges in our research.Therefore,this research built a model to classify the activities in different fineness degree and used the decision tree to split it.After taking some experiments with babies,we use the moving windows to split the data into many independent period samples,and label manually the samples with the activities.Limited by the frequency of the sensor,in experiment,this research mainly just uses the time domain features of 3-axle accelerations,resultant acceleration,the angles between axles and the resultant and the angular velocity.Decision tree will be used to classified the activities step by step.Supervised algorithm C4.5 decision tree and Support Vector Machines(SVMs)etc.were tested to be used for classification.Finally,this research chose SVM with RBF kernel.Because of the limited number of samples,K-folder cross validation is put into used in experiment to validate the classification algorithm.As the results,the possibility to recognize the static or moving is about 93% and the possibility to know if the toy is pushed or pulled is about 79%.The way this research recognizes children playing toys alone in the experiment,will be sent to the medical experts for the future research on the ASD early signs.