ABSTRACT:

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition that affects social communication, behavior, and sensory processing. Early detection of autism is crucial for effective intervention and support. This project presents an AI-powered Autism Detection System that integrates machine learning, facial emotion recognition, and eye tracking to assess autism likelihood based on user responses and real-time facial analysis.

The system employs a machine learning model trained on autism screening data, utilizing features such as emotion scores, eye engagement, and questionnaire-based responses to make predictions. Emotion detection is performed using a Convolutional Neural Network (CNN)-based model, which classifies facial expressions into different emotional states. Eye tracking is incorporated to measure user engagement by analyzing eye aspect ratios using dlib’s 68 facial landmarks. These extracted features are used to supplement a Voting Classifier model, consisting of Support Vector Machines (SVM), XGBoost, and Logistic Regression, trained to predict autism likelihood.
Additionally, a Flask-based interface is designed to facilitate real-time interaction with the system. The model takes live input from the webcam, processes facial emotions and eye behavior, and automatically generates responses to a standard autism screening questionnaire. The user’s responses, combined with real-time emotion and eye-tracking data, are passed to the trained machine learning model to determine the likelihood of autism.

Experimental results indicate that integrating computer vision-based behavioral analysis with traditional questionnaire-based approaches enhances prediction accuracy. The system also includes video recording capabilities for further analysis and improvement of behavioral assessments.
This research highlights the potential of AI-driven autism detection in healthcare, offering a non-invasive, automated, and accessible solution for preliminary autism screening
 
INTRODUCTION:

BACKGROUND OF AUTISM SPECTRUM DISORDER (ASD)

Autism Spectrum Disorder (ASD) is a neurodevelopmental condition that affects communication, social interaction, and behavior. It manifests in varying degrees, making diagnosis challenging without specialized assessments. Individuals with ASD often experience difficulties in understanding emotions, exhibit repetitive behaviors, and have unique sensory responses. Early detection is crucial for timely intervention, which can significantly improve an individual’s quality of life. However, traditional diagnosis relies on subjective evaluations, making it time-consuming and inaccessible for many in underserved areas.

THE NEED FOR AUTOMATED AUTISM DETECTION

The rising prevalence of autism has increased the demand for efficient and accessible diagnostic tools. Machine learning and computer vision enable automated detection by analyzing behavioral cues such as facial expressions and eye movements. These technologies provide objective and faster assessments, assisting healthcare professionals in early decision-making. Automated systems reduce the burden on specialists and allow large-scale screenings, improving early intervention efforts.

FACIAL EMOTION RECOGNITION AND EYE TRACKING IN AUTISM DETECTION

Individuals with ASD often struggle with emotional expression and perception. Machine learning models can analyze facial expressions in real-time to detect emotional processing difficulties. Additionally, eye-tracking provides insights into attention patterns, as individuals with autism often exhibit unique gaze behaviors. Traditional eye-tracking requires expensive equipment, but computer vision enables real-time tracking using standard webcams, making it practical for widespread use.

MACHINE LEARNING FOR AUTISM DETECTION

Machine learning models analyze facial expressions, eye movements, and behavioral responses to predict autism likelihood. These models learn from large datasets, allowing real-time and scalable assessments. Deep learning is employed for emotion recognition, while classical machine learning techniques analyze autism screening questionnaires. Combining multiple technologies enhances accuracy and reliability, bridging the gap between clinical expertise and automated diagnosis.
COMPUTER VISION AND DEEP LEARNING FOR AUTISM SCREENING
Computer vision enables real-time facial and eye movement analysis, making autism detection non-invasive and continuous. Open-source frameworks like OpenCV and Dlib assist in face detection, emotion classification, and eye-tracking. Deep learning, particularly Convolutional Neural Networks (CNNs), improves emotion recognition by identifying subtle facial patterns. This enhances the system’s ability to detect emotional and behavioral traits associated with ASD.

INTEGRATION OF SCREENING QUESTIONNAIRES AND DATA PROCESSING

Screening questionnaires remain essential for assessing social and communication skills. Machine learning automates questionnaire analysis, improving consistency and accuracy. Feature engineering, including data normalization and encoding, ensures high-quality input for predictive models. Addressing dataset imbalances through oversampling techniques improves model reliability, ensuring fair and accurate autism detection.

OPTIMIZING MODELS FOR HIGH ACCURACY

Hyperparameter tuning optimizes machine learning models like Support Vector Machines (SVM) and XGBoost, enhancing their performance. Techniques such as cross-validation and dropout layers prevent overfitting, ensuring better generalization. This optimization process is crucial for developing an accurate and efficient autism detection system capable of handling diverse user inputs.