SYSTEM AND METHOD FOR EARLY DETECTION OF AUTISM SPECTRUM DISORDER USING FACIAL RECOGNITION ENHANCED BY MACHINE LEARNING ALGORITHMS

Abstract

Autism is generally termed as pervasive disorder. The part ―Pervasive‖ infers that the ailment is severe. There have been many technical advances in detection of autism traits, Machine Learning is one among them. This paper presents an approach of Machine Learning algorithms and methods to distinguish autism traits. Although, diagnosis and analysis of this disease are carried out at several age, but its symptoms mostly seems in their infancy. Few classification algorithms are used to detect autism using facial recognition. It is not specifically designed or commonly employed for diagnosing or studying autism. However, it can be utilized as part of computer vision systems that analyze facial expressions or other facial features in autism research. Researchers in the field of autism have explored various computer vision techniques to study specific aspects of the condition, such as facial expressions, eye contact, and social interactions. These techniques often involve more sophisticated algorithms, including deep learning models like convolutional neural networks (CNNs). Autism is a neuro cognitive disorder among children and adults. Hence a diagnosis is required to solve the issue with better accuracy. This research focuses on different classification models of SVM, CNN(VGG16), Haar cascade using OpenCV. Using these models, a better accuracy can be obtained to detect autism spectrum disorder. Thus, a good accuracy of 93% is achieved from VGG16 compared to other classification models.

Specification

Description:ML is a field of study concerned with algorithms that learn from examples for solving a problem. The popular supervised ML algorithms are classification and regression trees (CART), logistic regression, linear discriminant analysis, boosting, SVM, ada-boost, random forest, Haar Cascade and Convolutional neural network VGG16. Then computer vision technique open-cv is used to extract features from images. RGB images are resized to 90 × 90 pixels and converted to grayscale images coded as an array with the label autistic (1) or non-autistic (0). Train test and cross-validation techniques are used to split data for train, validation, and test set. Then, apply several ML algorithms. , C , Claims:1. Machine learning algorithms for Classification (100): Support vector machine, Random Forest, Haar Cascade, Convolutional neural network VGG16
2. Classification algorithm applying (101) - Acquistion of images, Screening of face in the image, Feature extraction, Recognition designed using MLmodel.
3. Final Classification model for detecting autism (102) - analysing classification algorithm with good accuracy.
4. Prediction of autism (103) - designed to find whether the child is autistic by facial recognition

Documents