1. AUTISM SPECTRUM DISORDER CLASSIFICATION USING GATED
RECURRENT UNIT (GRU) NETWORKS BASED ON SCREENING
QUESTIONNAIRE DATA
PRESENTED BY,
920321106025 : PERIYASAMY J
920321106028 : REEMA SHERLIN S
920321106032 : SEBASTIN VINOTH T
Guided by,
Mr. R . SATHIS KUMAR AP/ECE
2. ABSTRACT
• The early detection of Autism Spectrum Disorder (ASD) is crucial for timely intervention and support.
This work presents a machine learning-based classification model using a Gated Recurrent Unit (GRU) to
predict the likelihood of ASD in children based on diagnostic questionnaire scores and demographic factors.
The dataset includes features such as responses to autism screening questions (A1 to A10), age, gender,
ethnicity, history of jaundice, autism in the family, and prior use of an autism assessment app.
• This allows the GRU to analyze the interrelationships between various screening responses, offering a more
nuanced prediction than traditional machine learning models. The performance of the model is evaluated
based on classification accuracy, precision, recall, and F1-score. Our model demonstrates effective
performance in distinguishing between ASD-positive and ASD-negative cases, highlighting the potential of
recurrent neural networks in supporting early autism diagnosis.
3. INTRODUCTION
• Autism Spectrum Disorder (ASD) is a complex neurodevelopmental condition that affects social
communication, behavior, and sensory responses.
• The prevalence of ASD has been increasing worldwide, leading to a growing need for early detection and
intervention.
• Early diagnosis is critical, as it enables timely support and therapies that can significantly improve
developmental outcomes. However, diagnosing ASD is often challenging due to the variability in symptoms,
which range from mild to severe across individuals.
• In recent years, machine learning techniques have gained attention for their potential to automate and
enhance the diagnostic process, providing quicker and more scalable solutions. By analyzing patterns in
questionnaire responses and demographic data, machine learning models can assist in identifying individuals
at risk for ASD, enabling earlier and more efficient detection.
4. LITERATURE SURVEY
s.No Title Author ,Year Inference
1 Spectral Brain Graph Neural
Network for Prediction of Anxiety
in Children with Autism Spectrum
Disorder
Peiyu Duan et al, 2024 Explored the feasibility of using spectral features to
predict the MASC-2 total scores. We proposed
SpectBGNN, a graph-based network.
2 A Data Mining Based Approach to
Predict Autism Spectrum Disorder
Considering Behavioral Attributes
Shaon Bhatta Shuvo et al,
2023
In this paper, the Random Forest classifier was used
for the prediction of ASD based on behavioral
attributes. We are satisfied with the results with the
overall accuracy of 0.96%.
3 Estimating Autism Severity in
Young Children From Speech
Signals Using a Deep Neural
Network
Weiping Ding et al, 2023 Built several Deep Neural Network (DNN)
algorithms to estimate ADOS scores and compared
their performance with Linear Regression and
Support Vector Regression (SVR) models.
4 Predicting the Symptom Severity
in Autism Spectrum Disorder
Based on EEG Metrics
Yangsong Zhang et al, 2023 Investigate whether the ASD symptom severity
could be predicted with electroencephalography
(EEG) metrics. Based on a publicly available dataset,
the EEG brain networks were constructed, and four
types of EEG metrics were calculated.
5. OBJECTIVES
Develop a GRU-Based Model
Enhance Diagnostic Efficiency
Improve Prediction Accuracy
Evaluate Model Performance
Contribute to AI-driven Healthcare
6. EXISTING SYSTEM
1 . Background Context
2 . Data Used
3 . Preprocessing Techniques
Label Encoding
Feature Scaling
Imputation
4 . Model Design
Input Layer >> GRU Layer >> Dropout Layer >> Dense Layer >> Loss Functions >> Optimizer >> Metrics
5. Training and Evaluation
7. DISADVANTAGES
Limited Accessibility of Early ASD Diagnosis
Manual Screening is Inefficient
Lack of Automated Tools for ASD Detection
Ineffective Utilization of Complex Data
8. PROPOSED SYSTEM
• The proposed system uses a Gated Recurrent Unit (GRU)-based neural network to classify Autism
Spectrum Disorder (ASD) based on questionnaire responses and demographic data.
• The dataset includes features such as screening questions, age, gender, and prior autism history,
which are preprocessed before being input into the GRU model.
• The GRU captures complex relationships between these features, enabling accurate classification of
ASD risk.
• The system is trained using a binary cross-entropy loss function and evaluated with metrics like
accuracy, precision, and recall. It is designed to be scalable for real-world applications, aiding early
ASD detection.
10. WORKING
Dataset Preparation
• Collect and preprocess data: Ensure that the dataset contains relevant features (e.g., questionnaire
responses, age, gender) and handle any missing values, scaling, and categorical encoding.
• Feature selection: Select the most important features for the model, ensuring both questionnaire and
demographic data are included.
Splitting the Dataset
• Train-test split: Divide the data into training, validation, and test sets for model learning, tuning, and
evaluation.
• Cross-validation: Implement k-fold cross-validation for robust model performance and to avoid
overfitting.
11. WORKING
GRU Model Architecture
• GRU Layer: Use a GRU layer to capture sequential dependencies and relationships in the input data,
making the model suitable for time-series-like data such as questionnaire responses.
• Dense Layers: Use fully connected layers to transform the GRU output into a binary prediction
(ASD or not).
Loss Function and Optimizer
• Binary Cross-Entropy Loss: Use this as the loss function since the task is binary classification,
allowing the model to learn from probability differences.
• Adam Optimizer: Choose the Adam optimizer for its efficiency in adapting learning rates and faster
convergence.
12. WORKING
Training the Model
• Training parameters: Train the model using parameters like batch size, epochs, and implement early stopping to
prevent overfitting.
• Model fitting: Fit the model on the training data and validate on the validation set to optimize performance.
Evaluation Metrics
• Accuracy, Precision, Recall: Evaluate the model’s performance using accuracy, precision, and recall, which are
particularly important for medical diagnostics.
• Confusion Matrix: Use a confusion matrix to analyze true positive, true negative, false positive, and false negative
rates.
13. TOOLS REQUIRED
• Hardware Tools
1. A PC with windows OS
• Software Tools
1. Python IDE 3.7.6
2. Software Packages
I. Numpy
II. Matplotlib
III. Pyramid-arima
14. DATASET
Dataset related to autism screening of adults that
contained 20 features to be utilised for further
analysis especially in determining influential autistic
traits and improving the classification of ASD cases.
In this dataset, we record ten behavioural features
(AQ-10-Child) plus ten individuals characteristics that
have proved to be effective in detecting the ASD
cases from controls in behaviour science.
19. ANALYSIS
Model Accuracy (%) Precision (%) Recall (%) F1-Score (%)
Proposed GRU Model 95 96 97 96
Logistic Regression 88 85 82 83
Random Forest 90 92 89 90
Support Vector
Machine
91 90 92 91
Decision Tree 85 80 78 79
20. CONCLUSION
• The implementation of a Gated Recurrent Unit (GRU) model for predicting Autism Spectrum Disorder
(ASD) represents a significant advancement in the early detection of this complex developmental condition.
By utilizing a diverse set of features derived from diagnostic questionnaires and demographic information,
this model effectively identifies patterns that may not be readily apparent in traditional analysis methods. The
ability of the GRU to process and interpret interrelated screening responses enables a more comprehensive
assessment of ASD risk, ultimately facilitating timely interventions and support for affected children and their
families.The model's performance metrics—accuracy, precision, recall, and F1-score—demonstrate its
robustness in classifying cases as either ASD-positive or ASD-negative.
21. REFERENCES
• S. Wang and N. C. Dvornek, "A Metamodel Structure For Regression Analysis: Application To Prediction Of
Autism Spectrum Disorder Severity," 2021 IEEE 18th International Symposium on Biomedical Imaging
(ISBI), Nice, France, 2021, pp. 1338-1341, doi: 10.1109/ISBI48211.2021.9434009.
• P. Duan et al., "Spectral Brain Graph Neural Network for Prediction of Anxiety in Children with Autism
Spectrum Disorder," 2024 IEEE International Symposium on Biomedical Imaging (ISBI), Athens, Greece,
2024, pp. 1-5, doi: 10.1109/ISBI56570.2024.10635753.
• S. B. Shuvo, J. Ghosh and A. S. Oyshi, "A Data Mining Based Approach to Predict Autism Spectrum
Disorder Considering Behavioral Attributes," 2019 10th International Conference on Computing,
Communication and Networking Technologies (ICCCNT), Kanpur, India, 2019, pp. 1-5, doi:
10.1109/ICCCNT45670.2019.8944905.
• Y. Zhang et al., "Predicting the Symptom Severity in Autism Spectrum Disorder Based on EEG Metrics,"
in IEEE Transactions on Neural Systems and Rehabilitation Engineering, vol. 30, pp. 1898-1907, 2022, doi:
10.1109/TNSRE.2022.3188564.
• M. Eni, I. Dinstein, M. Ilan, I. Menashe, G. Meiri and Y. Zigel, "Estimating Autism Severity in Young
Children From Speech Signals Using a Deep Neural Network," in IEEE Access, vol. 8, pp. 139489-139500,
2020, doi: 10.1109/ACCESS.2020.3012532.
