EARLY-STAGE AUTISM SPECTRUM DISORDER DETECTION USING MACHINE LEARNING

Abstract

This project needs an ML algorithm to diagnose Autism Spectrum Disorder early. ASD is difficult to minimize, however the business tries to decrease stimulation early in implantation. Four feature selection (FS) programs—Quantile Transformer, Power Transformer, Normalizer, Max Abs Scaler—will be tested against four common ASD datasets from children to females. The scaled datasets will be processed using ML algorithms including AdaBoost, RF, DT, KNN, GNB, LR, SVM, LDA, etc. Most ideal decision classifier and FS plan each gathering of same status are driven using numerical assessments. The polling classifier envisions ASD accompanying the best accuracy in infants, children, adolescents, and adults. An in-depth examination of feature significance using a four-determinant solving scheme highlights the need of adjusting ML approaches to predict ASD across age groups and helps healthcare providers find alternatives to camouflage. Distinguished to existent methods for early discovery of ASD, the submitted scheme acts well. To further help the flexibility and veracity of ASD discovery, an ensemble approach utilizing voting classifiers accompanying RF and AdaBoost completed 100% accuracy.