Soccer/footbal - section 4 - running the model without regularization.
Training set accuracy is 0.947, Test set accuracy is 0.91. Bias is not bad. Generalizability is good.
Why is this considered over-fitting?
HI @dds
As in this note book the difference between train accuracy and test accuracy is big, and the difference between the human accuracy and train accuracy is small …so if you deploy this model with these accuracy it will predict(the output) will be so bad as the model suffer from overfit and if you said that the difference between train accuracy and test accuracy is not big it is 0.3~ 0.4 may be I will not agree with you as train and test samples simply they came from same distribution but the real test(when you deploy the model in real live) in this case the test accuracy will be very small …so we doing regularization to avoid suffering from both(overfit & uderfit) by tuning the hyperparameter lambda
Please feel free to ask any questions,
Thanks,
Abdelrahman