Classification of Toddler, Child, Adolescent and Adult for Autism Spectrum Disorder Using Machine Learning Algorithm

• Published in: International Conference on Advanced Computing and Communication Systems (Online) Vol. 1; pp. 2444 - 2449
• Main Authors: Kavitha, V., Siva, R.
• Format: Conference Proceeding
• Language: English
• Published: IEEE 17.03.2023
• Subjects: Autism; Brain modeling; Classification; Convolutional Neural Network; Keywords-Autism spectrum disorder; Machine learning; Machine learning algorithms; Naïve Bayes; Particle Swarm Optimization (PSO); Pediatrics; Pre-processing; Predictive models; Sensitivity; Support Vector Machine; Support vector machines
• ISSN: 2575-7288
• DOI: 10.1109/ICACCS57279.2023.10112932

A neurodevelopmental syndrome called autism spectrum disorder (ASD) is frequently associated by sensory problems such an excessive or insufficient adaptability to noise, smells, or touch. Our daily lives are relying more and more on Machine Learning (ML). A person with an ASD has lifelong difficulties communicating and interacting socially. This condition begins in childhood and progresses throughout maturity. Therefore, this disorder completely affects a person's life. Reducing symptoms of autism spectrum disorders and improving quality of life for patients with autism, an early diagnosis is essential. Managing the subject's physical and mental health will be substantially helped by the early identification process. The manual screening procedure is more convenient and takes less time. The proposed work is based on an ASD detection mechanism that uses a convolutional neural network and a particle swarm optimization algorithm (PSO-CNN). Initial pre-processing removes missing information from the dataset. To increase the effectiveness of the suggested, we analyse the four underlying techniques, including the SVM, NB, LR, and PSOCNN with four different dataset types, including ASD Screening Data for Toddlers, Adults, Children, and Adolescents are used for the ASD Prediction analysis. Outcomes strongly indicate that PSO-CNN based prediction models perform more accurately and efficiently on all of these datasets of 99.1% after using various ML approaches with handling missing information.