IRJET - Prediction of Autistic Spectrum Disorder based on Behavioural Features using Machine Learning

International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 07 Issue: 04 | Apr 2020 www.irjet.net p-ISSN: 2395-0072
© 2020, IRJET | Impact Factor value: 7.34 | ISO 9001:2008 Certified Journal | Page 239
PREDICTION OF AUTISTIC SPECTRUM DISORDER BASED ON
BEHAVIOURAL FEATURES USING MACHINE LEARNING
Lakshmi B[1], Kala A[2]
[1] Student, Department of Information Technology, Sri Venkateswara College of Engineering, TamilNadu, India
[2]Asst. Professor, Department of Information Technology, Sri Venkateswara College of Engineering,
TamilNadu, India
---------------------------------------------------------------------***----------------------------------------------------------------------
Abstract - Autistic Spectrum Disorder is a neurodevelopmental disorder that affects a person’s interaction, communication,
learning skills and it is gaining momentum faster than ever. Detectingautismtraitsthroughscreeningtestsistime-consumingand
very expensive. With the advancement of machine learning and it’s algorithms, autismcanbepredictedatanearlystage. Although
there are a lot of studies using different techniques, these studies did not provide any definitive conclusion about the prediction of
autism in terms of different age groups. Therefore, this project aims at building a machine learning model that predicts the
disorder using supervised machine learning algorithms.
Key Words: ASD, Autism, Neural networks, machine learning, Adam
1. INTRODUCTION
Machine Learning is the field of study that gives computers the capability to learn without being explicitly
programmed. As it is evident from the name, it gives the computer that which makes it more similar to humans: The ability to
learn. Machine learning is actively being used today, perhaps in many moreplacesthanonewould expect.Thisprojectisabout
the application of machine learning in the field of health care, to identify ASD.
Autistic spectrum disorder is a neurodevelopmental disorder that affects a person’s interaction, communication,
learning skills and it is gaining its momentum faster than ever. Detecting autism traits through screening tests is time-
consuming and very expensive. With the advancement of machine learning and its algorithms, autism can be predicted at an
early stage.
Current explosion rate of autism around the world is numerousanditisincreasingata veryhighrate.Accordingtothe
WHO, about 1 out of every 160 children has ASD. Some people with this disorder can live independently, while others require
life-long care and support. Diagnosis of autism requires a significant amount of time and cost. Earlier detection of autism can
come to a great help by prescribing patients with proper medication at an early stage. It can prevent the patient’s condition
from deteriorating further and would help to reduce long term costs associated with delayed diagnosis.
Thus, a time-efficient, accurate and easy screening test tool isverymuchrequired whichwouldpredictautismtraitsin
an individual and identify whether or not they require comprehensive autism assessment. Therefore, this project aims at
building a machine learning model that predicts the disorder using Neural networks.
The objective of this work is to propose an autism prediction model using Neural Networks that could effectively
predict autism traits of an individual of any age. To be more precise, the focusofthiswork isondevelopinganautismscreening
application for predicting the ASD traits among people of age groups 4-11 years, 12-17 years and for people of age 18 and
more.
The rest of the paper is organized as follows. Section II discusses the related research done in this area previously.
Section III presents the research methodology. Section IV elaborates the detailed implementation of the proposedsystemand
the implemented system is evaluated in Section V. Finally, Section VI concludes the paper by highlighting the research
contributions, limitations and future plans to extend this work further.
2. LITERATURE SURVEY
This section briefly describes the works related to prediction techniques ofASD.Forexample,Kaziaimstopropose an
effective prediction model based on the ML technique and to develop a mobile application for predictingASDforpeopleofany
age. The autism prediction model was developed by merging Random Forest-CART (Classification and Regression Trees)and
Random Forest-Id3 (Iterative Dichotomiser 3). The proposed model was evaluated with AQ-10 dataset and 250 real datasets
collected from people with and without autistic traits. The evaluation results showed that the proposed prediction model

provides better results in terms of accuracy, specificity, sensitivity, precision and false positive rate (FPR) for both kinds of
datasets.[2]
Kayleigh provides a comprehensive review of 45 papers utilizing supervised machine learning in ASD, including
algorithms for classification and text analysis. In these 35 reviewed ASD researchstudies,themostcommonlyused supervised
machine learning algorithms were SVM and ADtree. Supervised machine learning algorithms were used to identify candidate
ASD genes, and to investigate obscure links between ASD and other domains.[3]
Milan used six personal characteristics age, sex, handedness, and three individual measuresof IQfrom851subjectsin
the Autism Brain Imaging Data Exchange (ABIDE) database to predict the model’s performance. While [1] Daniel analyzed an
eye movement dataset from a face recognition task, to classify children with and without ASD to obtainanaccuracyof88.51%.
[4]
D.P.Wall is The Autism Diagnostic Interview-Revised (ADI-R) is one of the most commonly used instruments for
assisting in the behavioural diagnosis of autism. The exam consists of 93 questions that must be answered by a care provider
within a focused session that often spans 2.5 hours. Machine learning techniques were used to study the complete sets of
answers to the ADI-R available at the AutismGeneticResearchExchange(AGRE)for891individualsdiagnosedwithautismand
75 individuals who did not meet the criteria for an autism diagnosis. The analysis showed that 7 of 93 items contained in the
ADI-R were sufficient to classify autism with 99.9% statistical accuracy.[5]
Bram is about Predicting if a child has Autism Spectrum Disorder proved possible by using developmental delay,
learning disabilities and speech or other language problems. Two methods wereused toidentifytheseverityofthe ASD.The1-
away method improved the accuracy from 54.1% to 90.2%, which is a significant increase. This and the fact that the severity
was based on input from just the caretakers of the children, prompts the need for further research in this matter.[6]
Wenbo identifies autism using Support Vector Machine (SVM) which provided accuracy up to 89% whereas [8] Jianbo
used Natural Language Processing (NLP) for autism detection based on information extracted frommedical formsofpotential
ASD patients. The proposed system achieves it an 83.4% accuracy and 91.1% recall, which is very promising.[7]
Chua combined a deep learningmethodwithSVMRFEtoimprovetheclassificationaccuracyofASDbasedonthe whole
ABIDE dataset. A total of 501 subjects with autism and 553 subjects with typical control across 17 sites were involved in the
study. The state-of-the-art average accuracy of 93.59%.[9] Anibal used deeplearningtechniquestoclassifyautismclassesusing
clinical datasets. [10]
From the literature review, it is evident that, though many types of research has been carried out in this field, the
researchers did not come to a decisive conclusion on using ML approach to predict autism for different age groups. Different
tools and methods were employed for autism screening tests, but none concentrated on different age groups.
3. RESEARCH METHODOLOGY
The research was carried out in four phases: Data Set collection, Data synthesis, Developing the prediction model,
Evaluating the predicted model. The phases are briefly discussed in the following subsections:
A. Data Set collection
To develop an effective predictive model, AQ-10 dataset was used whichconsistsofthreedifferentdatasetsbasedAQ-
10 screening tool questions. These three data sets contain data of age groups of 4-11 years (child), 12-17 years (adolescent)
and 18 years plus (adults). AQ-10 or Autism spectrum Quotienttool isusedtoidentify whetheranindividual shouldbereferred
for a comprehensive autism assessment. These questions mainly focus on domains like attention switching, communication,
imagination, and social interaction. Since the actual collectionofdata frompatients wouldbequitedifficult,thedata iscollected
from the UCI Machine Learning Repository as well, which is depicted in Fig 1.

Fig 1. Sample data set from the UCI Machine Learning Repository
B. Data synthesis
The collected data were synthesized to remove irrelevant features. For example, ID was irrelevant to develop a
prediction model, hence it was removed. Further, unnecessary fields were deleted using pandas. This is done in order to
increase the accuracy in classification. Summary of synthesized datasets is shown in Table 1.
Table 1. Summary of the chosen dataset
Age Group Total Cleaned Instances % of Male-Female Average Age
4-11 years 248 70.16% male,
29.84% female
6.43 years
12-16 years 98 50% male,
50% female
14.13 years
18 and more 608 52.7% male,
47.3% female
29.63 years
C. Developing the prediction model
To generate a prediction of autism traits, algorithmshadbeendevelopedandtheiraccuracywastested.Afterattaining
results from various types of machine learning techniques such as Linear regression,SVM,NaiveBayes;Neural networkswere
found to be highly feasible with higher accuracy than the other algorithms. So, Neural Networks was proposed for
implementing the ASD predictive system. Further modifications were made to the algorithm to get better results.

Fig 2. Development and Prediction of the model
D. Evaluating the predicted model
The model is tested with data that has been trained with the help of neural networks. This is used in fine-tuning the
prediction. Of all the data taken from the UCI Machine Learning Repository, 80% of it is used for training the model and the
remaining 20% of the data is used for testing. Testing helps us to fine-tune the model further to increase the accuracy in
prediction. With this, a 90% accuracy was achieved.
Fig 3. Evaluation and accuracy results

4. IMPLEMENTATION OF PROPOSED SYSTEM
The model is trained using the data set that has been pre-processed.Thisisusedforproper predictionlater.Hence,the
data set is used to train multiple models. A neural network is a series of algorithms that endeavours to recognize underlying
relationships in a set of data through a process that mimics the way the human brain operates. In this sense, neural networks
refer to systems of neurons, either organic or artificial in nature. Neural networks can adapt to changing input;sothenetwork
generates the best possible result without needing to redesign the output criteria.
The neuron of the proposed system is a combination of a linear and a nonlinear function that takes up vectors
comprising the various attributes that are defined in the AQ-10 dataset. The importance of various attributes is defined in the
weight function. This linear combination can be depicted as:
f(x1, x2) = w1x1 + w2x2 --- (1)
Equation 1. A linear function of the neural network
The nonlinear part of the model also called the activation function, is represented using the ReLU function. A neural
network without an activation function is essentially just a linearregressionmodel.The activationfunctiondoesthe non-linear
transformation to the input making it capable of learning and performing more complex tasks.
ReLU(x) = max(x,0) --- (2)
Equation 2. The activation function of the neural network
Thus, the model can be represented as:
f(x1, x2) = max(0, w1x1+w2x2) --- (3)
Equation 3. Proposed neural network model
Adam algorithm is used for this neural network.
Fig 4. Basic Adam algorithm functionality
5. EVALUATION OF PROPOSED SYSTEM
The prediction is the actual accurate identification of the autism data basedontheinputgiven.Owingtothedata given
the model is trained in a better way. More data, more fine-tuning. Hence, bigger datasets give more accuracy.Theadvantageof
the usage of neural networks for prediction is that they are able to learn from examples only and that after their learning is
finished, they are able to catch hidden and strongly non-linear dependencies, even when there is significant noise in the
training set. The disadvantage is that neural networks can learn the dependency valid in a certain period only. The error of
prediction cannot be generally estimated. However, the accuracy was close to 90%. This was implemented using Python with
Keras data processing package. With multiple epochs andbatchprocessing,theaccuracywasfoundtobecloseto90%.Thiscan
still be fine-tuned, which is covered in future work.

Fig 5. Accuracy of the proposed model
6. CONCLUSION AND FUTURE WORK
Autism is quite common, and with the results, one can find out which is the major contributing factortowardsautism.
Since the data set is quite comprehensive in terms of the factors, one can easily scrutinizesuchpregnantmothersandtakecare
in the initial stages. Also, this will help the health care providers to split the funding and care accordingly.
The primary limitation of the study is the lack of sufficiently largedata totrainthemodel.Anotherlimitationisthatthe
screening application is not designed for the age group below 3 years as open-source data was not available. Our future work
will focus on collecting more data from various sources to improve the accuracy of the proposed system to take it to a higher
level.
REFERENCES
[1] Kazi Shahrukh Omar, Prodipta Mondal, Nabila Shahnaz Khan, Md. Rezaul Karim Rizvi, Md Nazrul Islam, “A Machine
Learning Approach to Predict Autism Spectrum Disorder”,2019 International Conference on Electrical, Computer and
Communication Engineering (ECCE), 7-9 February, 2019
[2] Kayleigh K. Hyde, Marlena N. Novack, Nicholas LaHaye, Chelsea Parlett-Pelleriti, Raymond Anden, Dennis R. Dixon, Erik
Linstead, “Applications of Supervised Machine LearninginAutismSpectrumDisorderResearch:a Review”,ReviewJournal
of Autism and Developmental Disorders (2019) 6:128–146, https://doi.org/10.1007/s40489-019-00158-x
[3] https://ctb.ku.edu/en/table-of-contents/evaluate/evaluate-community-interventions/information-gathering-
synthesis/main
[4] D. P. Wall, R. Dally, R. Luyster, J.-Y. Jung, and T. F. DeLuca, “Use of artificial intelligence to shorten the behavioral diagnosis
of autism,” PloS one, vol. 7, no. 8, p. e43855, 2012.
[5] Bram van den Bekerom, “Using Machine Learning for Detection of Autism Spectrum Disorder”, 2017
[6] Wenbo Liu, Ming Li, and Li Yi, “Identifying Children with Autism Spectrum Disorder Based on Their Face Processing
Abnormality: A Machine Learning Framework”, Autism Research 00: 00–00, 2016
[7] Jianbo Yuan, Chester Holtz, Tristram Smith, Jiebo Luo, “Autism spectrum disorder detection from semi-structured and
unstructured medical data”, EURASIP Journal onBioinformaticsandSystemsBiology(2017)2017:3DOI10.1186/s13637-
017-0057-1

[8] Daniel Bone, Matthew S. Goodwin, Matthew P. Black, Chi-Chun Lee, Kartik Audhkhasi, Shrikanth Narayanan, “Applying
Machine Learning to Facilitate Autism Diagnostics:PitfallsandPromises”, Journal ofAutismandDevelopmental Disorders,
May 2015, Volume 45, Issue 5, pp 1121–1136
[9] Chua Wang, Zhiyong Xiao, Baoyu Wang, Jianhua Wu, “IdentificationofAutismBasedon SVM-RFEandStackedSparse Auto-
Encoder”, IEEE Access ISSN-2169-3536, 21 August, 2019
[10] Jared A. Nielsen, Brandon A. Zielinski, P. Thomas Fletcher, Andrew L. Alexander, Nicholas Lange, Erin D. Bigler, Janet E.
Lainhart and Jeffrey S. Anderson, “Multisite functional connectivity MRI classification of autism: ABIDE”, Front. Hum.
Neurosci., 7 (September)(2013), pp. 1-12
[11] Plitt M., Barnes K.A., Martin A. “Functional connectivity classification of autism identifies highly predictive brain features
but falls short of biomarker standards” NeuroImage: Clinical, 7 (2015), pp. 359-366
[12] Anibal Sólon Alexandre Rosa Franco, R. Cameron Craddock,AugustoBuchweitz,FelipeMeneguzz,“Identificationof autism
spectrum disorder using deep learning and the ABIDE dataset”, NeuroImage: Clinical Volume 17, 2018, Pages 16-23,
https://doi.org/10.1016/j.nicl.2017.08.017
[13] Milan N. Parikh, Hailong Li1 and Lili He, “Enhancing Diagnosis of Autism With Optimized Machine Learning Models and
Personal Characteristic Data”, Front. Comput. Neurosci.,15 February 2019
[14] Dua, D. and Graff, C. (2019). UCI Machine Learning Repository [http://archive.ics.uci.edu/ml]. Irvine, CA: University of
California, School of Information and Computer Science.
[15] Yuji Roh, Geon Heo, Steven Euijong Whang, “A Survey on Data Collection for Machine Learning A Big Data - AI Integration
Perspective”, IEEE, August 2019
[16] Ibrahim M. Nasser, Mohammed O. Al-Shawwa, Samy S. Abu-Naser, “Artificial Neural Network for Diagnose Autism
Spectrum Disorder”, International Journal of Academic Information Systems Research (IJAISR), February 2019

IRJET - Prediction of Autistic Spectrum Disorder based on Behavioural Features using Machine Learning

More Related Content

What's hot (20)

Similar to IRJET - Prediction of Autistic Spectrum Disorder based on Behavioural Features using Machine Learning (20)

More from IRJET Journal (20)

Recently uploaded (20)

IRJET - Prediction of Autistic Spectrum Disorder based on Behavioural Features using Machine Learning