I’m doing forecasting on covid 19 outbreak predicitions for several machine learning models, such as svm, ann, random forest, and knn. i’m quite confused on how to compare which is which is the best accuracy performance model with the train and test accuracy result, mean-squared error, and also r-squared. and another one question, what usually the units preferred for train and test accuracy result? whether it is in percent/whole number/decimal point

to get a clear explanation on the problems above