I am facing a problem, i have a point error dataset of 74 rows in which 3 are my inputs and 16 are output features. Well when I trained my model using polynomial regression, if I increased the degree I got good results in train r2 and test r2 and this is overfitting. when I used DNN model with complex hidden layer I got good train r2 but test r2 fails so what you suggest to get good train and test r2, either I need to change my model as when I tuned my polynomial model the results not improved or we have to collect more data ?