Predicting Autism Spectrum Disorder (ASD) meltdown using Fuzzy Semi-Supervised Learning with NNRW
Sara Karim, Nazina Akter, Muhammed J. A. Patwary
Abstract
Autism Spectrum Condition (ASD) is a notable psychological disorder that affects a human’s ability to communicate socially. The need of early diagnosis prompted researchers’ attention to the usage of various machine learning-based approaches. A number of supervised and unsupervised learning methods from the disciplines of machine learning were used to enhance the precision of the Autism prediction system. In supervised learning systems, only labeled data is utilized to build a classification model, but gathering enough classified data is time-consuming and usually involves field specialist’s assistance. However, in various real-world circumstances, unlabeled samples are easily available. By integrating large volumes of labeled and unlabeled data to produce a stronger classification model, Semi-Supervised Learning (SSL) solves this problem better than all other Machine learning methods. The study proposes a novel fuzziness-based semi-supervised learning strategy for ASD meltdown prediction that employs mislabeled data in conjunction with training labeled data to increase the model’s reliability. The prediction performance utilizing this method on the dataset demonstrated significant advancements to increasing the classifier’s results in comparison to other classification methods such as Fuzzy MinMax Classifier, Naive bayes, Fuzzy Data Mining, ZeroR, Random tree, and others.