Table 8.

A comparison of our proposed models with previous work.

Refs	Dataset	Models	Accuracy	Limitations	Strengths
1	Children 1054 instances along with 18 attributes	LR, SVM, KNN, RF	LR : 93.15% NB:94.79% SVM:93.84% KNN:90.52% RFC: 81.52%	The dataset was compiled from primarily autism-based collections, as a result of which there was quite a significant imbalance, in favor of the ASD class	Used forward feature selection Trained and tested five ML models
5	2009 features records (toddlers, children, adults)	SVM, Glmboost and adaboost classification	85.10%, 97%, 98%	Constrained sample size/data set	Metrics based on brain activity used for prediction of ASD
7	Children have 292 instances	Naïve Bayes, SVM, LR, RF, CNN, NN	97.53%, 9 6.30%, 96.88%	Does not predict the severity of ASD. Conditions used for identification of ASD which might not always necessarily translate to a case of ASD	Makes use of six different ML based classification methods and obtained high accuracy
Proposed models	Children Records: > 200 Attributes: 22 Adults Records: > 700 Attributes: 21	SVM, LR	Children SVM: 99% LR: 98% Adults SVM: 81% LR: 78%	Size of children dataset is very small	Comparable accuracy has been obtained Two different datasets of children and adults have been processed State of the art FL technique has been applied for ASD prediction