Autism Detection Using HAAR Cascade Machine Learning Algorithm

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1417
Autism Detection Using HAAR Cascade

Machine Learning Algorithm
Lakshmiprabha1; Shivam Patil2; Sumit Kakad3; Sanket Patil4
1
Associate Professor, Department of Electronic and Telecommunication Engineering,
D. Y. Patil Internation University, Pune, India
2;3;4
Students, Department of Electronic and Telecommunication Engineering,
Dr. D. Y. Patil Institute of Management and Research, Pune, India
Publication Date: 2025/03/31
Abstract: Autism Spectrum Disorder It refers to a big spectrum of conditions that influence social interactions,
communication skills, and repetitive behaviors. Traditionally, ASD diagnosis relies on behavioural observations, but there is
increasing interest in leveraging technology for earlier detection. This project explores using the HAAR Cascade algorithm,
typically employed for object detection like facial recognition, to identify different ASD types. We concentrated on four
categories: Asperger Syndrome, Childhood Disintegrative Disorder, Pervasive Developmental Disorder-Not Otherwise
Specified (PDD-NOS), and Classic Autism are all conditions that fall under the umbrella of developmental disorders. Our
approach involved training separate HAAR Cascade models for each type using meticulously labelled images. Positive
samples highlighted features associated with each condition, while negative samples included unrelated facial characteristics.
The system analyzes new images to classify the type of ASD or indicate no detection if relevant features are absent. Although
HAAR Cascade is generally used for simpler tasks, this project aimed to assess its capability in this complex application.
The success of our system heavily depended on the quality of the training data and the precision of feature identification by
each model. This project is an initial exploration into using HAAR Cascade for ASD detection, suggesting that more
advanced techniques, such as deep learning, may be necessary for improved accuracy. Our findings could inform future
research, potentially leading to more effective combined methods.
Keywords: Autism Spectrum Disorder (ASD), ASD Detection, HAAR Cascade, Object Detection, Asperger Syndrome, Childhood
Disintegrative Disorder, Pervasive Developmental Disorder-Not Otherwise Specified (PDD-NOS), Classic Autism, Facial Features,
Image Analysis, Classification, Training Data, Positive Examples, Negative Examples, Variability in Facial Features, Deep
Learning, Future Research, Combined Methods, Diagnostic Accuracy.
How to Cite: Lakshmiprabha; Shivam Patil; Sumit Kakad; Sanket Patil (2025). Autism Detection Using HAAR Cascade Machine
Learning Algorithm. International Journal of Innovative Science and Research Technology, 10(3), 1636-1641.
https://doi.org/10.38124/ijisrt/25mar1417
I. INTRODUCTION This project proposes a computer vision-based approach for

detecting autism by analyzing facial expressions in real-time
Autism Spectrum Disorder (ASD) is a multifaceted using Python’s OpenCV library and Haar Cascade classifiers.
neurodevelopmental affliction characterized by difficulties in The system processes live video input from a camera,
interaction and behavioral patterns. As per the information analyzes facial features, and classifies four different types of
from the Centers for Disease Control and Prevention (CDC) autism based on key expression patterns. By employing
Shows that around 1 in 44 children in the United States are image preprocessing techniques and a trained classifier, the
identified with autism, which aligns closely with global system aims to provide accurate and reliable detection,
prevalence trends. Due to the spectrum's diverse nature, contributing to early diagnosis.
individuals display distinctive traits, making early
identification and intervention challenging. However, early This project highlights the importance of using
diagnosis is vital, as it allows for timely support that can noninvasive, facial recognition techniques to identify
greatly enhance developmental progress. including potential indicators of autism, offering an accessible tool for
behavioral therapies, educational support, and family healthcare professionals and caregivers. The research
counseling. However, diagnosing autism is challenging due advances the use of machine learning and computer vision in
to the subjective nature of behavioral assessments and the autism detection, paving the way for further developments in
need for trained professionals to interpret complex cues. assistive technologies.
Consequently, there is a growing demand for objective,
technology-driven methods for early detection and diagnosis.
IJISRT25MAR1417 www.ijisrt.com 1636

II. LITERATURE REVIEW  Objective
 Characteristics of Autism Spectrum Disorder (ASD):  Develop a HAAR Cascade-based Detection System:
Create a system using HAAR Cascade classifiers to detect
 Ongoing challenges with communication and interaction ASD-related facial features and distinguish between
that occur across different contexts. specific ASD types, this includes Asperger Syndrome,
 The presence of repetitive behaviors or a pronounced Childhood Disintegrative Disorder, Pervasive
resistance to alterations in daily routines. Developmental Disorder-Not Otherwise Specified (PDD-
 Symptoms generally become apparent during early NOS), and Classic Autism.
childhood, often manifesting within the very first years  Train and Validate Classifiers: Train individual HAAR
of life. Cascade models for each ASD subtype using labeled
positive and negative image samples, ensuring reliable
 Header: feature identification for each condition.
Brain Growth, Structure, and Connectivity in Autism  Evaluate Feasibility of HAAR Cascade for Complex
Spectrum Disorder (ASD) Authors: Bartholomeusz HH, Detection: Assess the effectiveness of HAAR Cascade,
Courchesne E, Karns CM Source: Neuropediatrics (October traditionally used for simpler tasks, in identifying
2002, Volume 33, Issue 5, pp. 239–241). nuanced facial features related to ASD, and determine its
potential in this more complex application.
 Info:  Contribute Foundational Research for Future
Research utilizing head circumference measurements in Approaches: Establish baseline findings on HAAR
infants and young children diagnosed with autism has shed Cascade’s applicability in ASD detection to inform and
light on abnormal patterns of early brain development. Head inspire future studies, potentially combining this approach
circumference, a reliable indicator of brain size in early with deep learning for improved accuracy.
postnatal stages, has been pivotal in identifying both  Enhance Early Diagnostic Accessibility: Explore the
accelerated and reduced growth tendencies in Autism potential of an image-based system to support accessible
Spectrum Disorder (ASD). Numerous studies have reported early diagnosis, especially in under-resourced areas,
unusually large head circumferences in children with ASD, ultimately aiding in timely intervention for individuals
particularly around the time when symptoms first become with ASD.
noticeable. These findings underscore significant deviations
in brain growth trajectories that are associated with ASD. III. BACKGROUND
 Header: Autism Spectrum Disorder (ASD) is a A multifaceted

Nutrition and the Development of Autism Spectrum neurodevelopmental disorder characterized by challenges in
Disorder (ASD) Authors: Thommen, E., Suarez, M., Guidetti, connecting with others, navigating social interactions, and
M., Guidoux, A., Rogé, B., & Reilly, J. S. Source: engaging in repetitive behavioral patterns. As a spectrum
Understanding Emotions in Children with Autism: Crossed disorder, its severity and expression differ significantly
Perspectives According to Tasks, Enfance, 2010; Volume 3, among individuals. Timely diagnosis is crucial to enable
pp. 319–337. interventions that can greatly enhance developmental
progress and quality of life for those affected. Nonetheless,
 Info: conventional diagnostic techniques predominantly depend on
Both parents and doctors working with children behavioral evaluations, which are inherently subjective and
diagnosed with autism spectrum disorder (ASD) have can vary based on the expertise and perspective of the
observed a potential connection between dietary habits and evaluator.
the intensity or frequency of symptoms. Several theories have
been proposed to explain this relationship, including the One promising approach to improving diagnosis is the
involvement of food additives or specific dietary components use of machine learning and computer vision techniques,
in the development of ASD. Recent findings from animal particularly those focused on facial expression analysis.
research have revealed that propionic acid (PA), a dietary Studies have shown that individuals having Autism Spectrum
short-chain fatty acid commonly used as a food additive, can Disorder (ASD) often exhibit distinct facial expression
trigger neuroinflammatory responses and behavioral changes patterns and emotional responses compared to neurotypical
in rats that closely resemble symptoms associated with ASD. individuals. These differences, though subtle, can be detected
by automated systems designed to analyze facial features and
expressions, providing an objective and non-invasive
diagnostic tool.

Fig 1 Here we can see the Frame Selection Algo in Pictorial Form also here we Crop the Face of Child for Further Process.
 HAAR Cascade Classifier for Object Detection In the context of ASD detection, the HAAR Cascade
The HAAR Cascade Classifier is a learning-based Classifier can be extended beyond simple face detection to
method used for object spotting in images and video streams. identify specific facial features and expressions that may
Originally developed for face detection, HAAR Cascade uses indicate ASD. By training the classifier with a labelled dataset
simple rectangular features to identify objects based on the of facial images from individuals with different types of
presence of particular patterns in the image. It works by autism, the system can learn to recognize facial expression
scanning an image at multiple scales and detecting faces by patterns corresponding to Asperger Syndrome, Childhood
analyzing the image at various resolutions. The classifier is Disintegrative Disorder, PDD-NOS, and Classic Autism.
trained using a dataset of positive (faces) and negative (non-
faces) examples, learning to distinguish between these two by  OpenCV and HAAR Cascade for Autism Detection
detecting the patterns that differentiate them. In addition to facial expression analysis, other important
technologies used in this project include the OpenCV library,
which provides a comprehensive set of tools for.
Fig 2 This Fig Explains the Steps Involved in Identifying Autism Spectrum Disorders through Facial Recognition and
Classification Algorithms.

Real-time computer vision tasks, such as image IV. METHODOLOGY AND DESIGN
processing, feature extraction, and object detection. OpenCV
helps process video input, detect faces in real-time, and The proposed methodology involves several key steps.
analyze facial expressions by tracking facial landmarks. By First, a dataset of facial images representing individuals with
combining HAAR Cascade with these technologies, A system various types of autism will be collected. This dataset This
can be designed to identify and categorize autism-related will form the basis for training the HAAR Cascade classifier.
expressions with greater speed, consistency, and potentially Various image preprocessing techniques will be employed to
higher accuracy compared to conventional methods. improve the quality of the input data. ensuring that the
classifier can accurately detect and analyze facial features.
This theoretical background highlights the foundation
for applying computer vision techniques like HAAR Cascade Once the model is trained, the system will utilize real-
Classifiers to ASD detection, providing the necessary time video input from a camera to capture facial expressions.
framework for this project's development and offering a The trained classifier will process each frame to identify and
pathway for further advancements in early diagnosis and classify the detected expressions, providing insights into the
intervention. potential indicators of autism.
 Facial Morphology with HAAR Cascade  Planning

This project aims to create a system utilizing Python's The planning phase was critical in defining the project's
OpenCV library and the HAAR Cascade Classifier to core objectives and setting a clear direction for the
examine facial expressions and categorize four specific forms development process. Key goals included the need for real-
of autism: Asperger Syndrome, Childhood Disintegrative time data updates, a highly responsive user interface, and
Disorder, Pervasive Developmental Disorder-Not Otherwise secure authentication mechanisms. The goal behind this
Specified (PDD-NOS), and Classic Autism. By training the project is to develop a system that uses the HAAR Cascade
classifier with a dataset of facial images labelled according to algorithm to detect autism-related traits in individuals based
these types, the system aims to detect and classify facial on facial expressions or behavioural patterns. The system will
expressions in real-time video inputs, offering a valuable, analyze images or video streams in real time, providing
non-invasive diagnostic tool for healthcare professionals. insights or alerts for caregivers or medical professionals.
 Problem Definition To meet these goals, it was crucial to select appropriate

The project "Autism Detection Using HAAR Cascade technologies and tools. The decision to use Next.js for the
Machine Learning Algorithm" aims to address the challenge front end and Convex for the back end reflected a focus on
of identifying autism spectrum disorder (ASD) characteristics real-time performance and scalability. Additionally, robust
through facial feature analysis. Autism Spectrum Disorder authentication was identified as a key requirement, leading to
encompasses a variety of developmental conditions that the selection of OAuth2 to enable secure logins through third-
impact social interaction, communication, and behavioral party providers such as Google.
patterns, and its early detection remains a critical need.
Traditional diagnosis often depends on behavioral  Design
assessments conducted by specialists, which can be time- In the design phase, the platform was modeled with a
consuming and may delay early intervention. This project modular architecture. This design allowed for the
explores the feasibility of using the HAAR Cascade machine independent development and deployment of core
learning algorithm to detect specific facial features associated components, such as the frontend, backend, authentication,
with four ASD subtypes: Asperger Syndrome, Childhood and database systems. The modular approach not only
Disintegrative Disorder, Pervasive Developmental Disorder- enhanced flexibility but also ensured that each component
Not Otherwise Specified (PDD-NOS), and Classic Autism could be developed and tested individually, reducing the risk
are developmental conditions that fall under Autism of integration issues later in the development cycle.
Spectrum Disorder (ASD). This project aims to adapt and
train the HAAR Cascade classifier, a tool typically used for The machine learning module, integrated with OpenCV,
object detection in images, to identify facial patterns and will employ the HAAR Cascade classifier to detect facial
features associated with these specific ASD subtypes. The attributes and features indicative of autism. This classifier
core goal is to create a prototype system capable of analyzing will undergo training using a meticulously curated collection
facial images and, using HAAR Cascade-based models, of facial images, ensuring its capability to recognize autism-
either classify them under an ASD subtype or indicate the related traits promptly and with high efficiency. The system
absence of ASD-related characteristics. However, due to the will preprocess the data (e.g., converting images to grayscale
complexity of ASD and the subtle differences in facial and detecting facial regions) to optimize detection speed and
features, this initiative serves as a foundational exploration accuracy. The database (ikely MongoDB) will store user
into ASD detection through technology. It provides a basis for profiles, session data, and detailed reports, offering an easily
future advancements, such as incorporating deep learning accessible way to manage and retrieve data.
models, to enhance detection accuracy and reliability.

Fig 3 Block Diagram to Show Process in Simple Manner.
V. FUTURE SCOPE make ASD detection more accessible, allowing caregivers

or educators to conduct preliminary screenings. A remote
The future scope of the project "Autism Detection Using application could support rural and underserved areas
HAAR Cascade" could include several promising with limited access to healthcare services.
advancements and extensions to enhance its utility, accuracy,
and scalability. Here are some key areas for future  Multi-stage Diagnostic Pipeline: Creating a multi-stage
exploration: diagnostic tool could allow the system to flag potential
cases and guide them for further in-depth assessment by
 Incorporating Deep Learning Models: Although HAAR specialists. This staged approach could be beneficial in
Cascade offers reliability in fundamental feature clinical settings, where early-stage screening can lead to a
detection, advanced deep learning frameworks, such as more detailed follow-up evaluation.
Convolutional Neural Networks (CNNs) and specialized
models like ResNet, are better equipped to identify  Integration with Clinical and Educational Tools:
intricate and subtle patterns with greater precision. Partnering with healthcare providers and educational
Combining HAAR Cascade with deep learning could institutions to incorporate this tool into existing diagnostic
improve the system's ability to differentiate ASD-related workflows could support a more holistic diagnostic
facial features from typical features. approach, combining facial analysis with traditional
behavioral assessments.
 Development of a Hybrid Model: Creating a hybrid
system that combines HAAR Cascade for initial facial  Investigating Alternative Learning Methods: Future
region detection with deep learning for in-depth analysis studies might consider utilizing other learning
could enhance detection accuracy. This combination approaches, like Support Vector Machines (SVMs) or
could refine feature extraction and improve classification decision trees, to enhance or fine-tune the classification
across different ASD subtypes. process. Testing these methodologies could aid in
identifying the most effective model tailored to specific
 Expanding Dataset Diversity and Size: Gathering a larger, ASD subtypes.
more diverse dataset with a variety of facial expressions,
ages, and ethnic backgrounds could help the system  Ethical and Privacy Considerations: As the project scales,
generalize better and perform more accurately across addressing ethical considerations, such as data privacy,
different populations. Including more labeled examples informed consent, and appropriate usage, becomes
for each ASD subtype would also improve model training. critical. Implementing strict privacy standards and
transparency about data handling can build trust with
 Feature Expansion Beyond Facial Analysis: While this users and ensure responsible technology deployment.
project focuses on facial features, incorporating other
ASD indicators—such as gaze patterns, facial  Continuous Learning System: Implementing a system that
expressions, or head movements—could enhance early learns continuously from new data, using semi-supervised
detection capabilities. Future models might analyze video or unsupervised learning techniques, could allow the
data to capture behavioral indicators that complement model to improve over time. This adaptive learning could
static image analysis. make the system more robust and responsive to subtle
changes in ASD-related research.
 Mobile and Cloud Deployment for Remote Screening:
Developing a mobile app or cloud-based platform could

Implementing these future strategies has the potential to [4]. Le Couteur, A., Lord, C., & Rutter, M. (2003).
develop a robust, user-friendly, and precise ASD screening Automated observation of the child’s social behavior
system. Such a tool could play a critical role in facilitating using computers. Journal of Child Psychology and
early diagnosis and providing essential support for Psychiatry, 44(4), 513521.
individuals on the autism spectrum. [5]. Picard, R. W. (1997). Affective Computing.
Cambridge, MA: MIT Press.
VI. ACKNOWLEDGEMENT [6]. Poudel, U. P., et al. (2020). "Facial Expression
Recognition: A Survey." International Journal of
We extend our sincere gratitude to D Y Patil Computer Applications, 975, 17.
International University for providing the essential resources [7]. Sutherland, R. et al. (2014). "Technologies for the
and support for this research. We sincerely thank our mentor Identification of Autism Spectrum Disorders in Young
for their invaluable guidance and our peers for their support Children." Pediatrics, 133(1), 148153.
and insightful feedback.
VII. CONCLUSION
In conclusion, the "Autism Detection Using HAAR

Cascade" project represents a significant first step in
exploring the potential for technology-assisted early
screening of autism spectrum disorder (ASD). By adapting
the HAAR Cascade algorithm—commonly used in simpler
object detection tasks—this project set out to determine if
facial features associated with specific ASD subtypes could
be identified. Through separate models trained for Asperger
Syndrome, Childhood Disintegrative Disorder, PDD-NOS,
and Classic Autism, the project aimed to classify ASD-related
characteristics and distinguish them from non-ASD traits.
Despite the inherent challenges, such as the subtlety of

ASD-related facial features and the limitations of HAAR
Cascade in complex feature differentiation, this project
demonstrated the feasibility of an image-based approach to
ASD screening. The results underline the importance of high-
quality training data and suggest that more advanced machine
learning models, such as deep learning, may be needed to
enhance accuracy and reliability.
This project opens the door for further research and

improvement, including the use of hybrid models, expanded
datasets, and the integration of complementary behavioral
indicators. With continued development and ethical
consideration, such technology holds promise for aiding early
ASD detection, making preliminary screening more
accessible, and potentially supporting earlier intervention.
Ultimately, this work provides a foundation upon which more
sophisticated, effective diagnostic tools for ASD can be built
in the future.
REFERENCES
[1]. American Psychiatric Association. (2013). Diagnostic

and Statistical Manual of Mental Disorders (5th ed.).
Arlington, VA: American Psychiatric Publishing.
[2]. Centers for Disease Control and Prevention (CDC).
(2021). Data & Statistics on Autism Spectrum
Disorder. Retrieved from [CDC
website](https://www.cdc.gov/ncbddd/autism/data.ht
ml).
[3]. Ekman, P. (1999). Basic Emotions. In T. Dalgleish &
M. J. Power (Eds.), Handbook of Cognition and
Emotion (pp. 4580). New York: Wiley.

Autism Detection Using HAAR Cascade Machine Learning Algorithm

Uploaded by

Autism Detection Using HAAR Cascade Machine Learning Algorithm

Uploaded by

Volume 10, Issue 3, March – 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar1417

Autism Detection Using HAAR Cascade

Publication Date: 2025/03/31

I. INTRODUCTION This project proposes a computer vision-based approach for

IJISRT25MAR1417 www.ijisrt.com 1636

 Header: Autism Spectrum Disorder (ASD) is a A multifaceted

IJISRT25MAR1417 www.ijisrt.com 1637

IJISRT25MAR1417 www.ijisrt.com 1638

 Facial Morphology with HAAR Cascade  Planning

 Problem Definition To meet these goals, it was crucial to select appropriate

IJISRT25MAR1417 www.ijisrt.com 1639

Fig 3 Block Diagram to Show Process in Simple Manner.

V. FUTURE SCOPE make ASD detection more accessible, allowing caregivers

IJISRT25MAR1417 www.ijisrt.com 1640

In conclusion, the "Autism Detection Using HAAR

Despite the inherent challenges, such as the subtlety of

This project opens the door for further research and

[1]. American Psychiatric Association. (2013). Diagnostic

IJISRT25MAR1417 www.ijisrt.com 1641

You might also like