Your model scores 99 % accuracy on training data but 72 % on validation data. What is happening, how do you confirm it, and what would you try?
Show the explanation
Model answer. A large gap between training and validation performance indicates high variance (overfitting): the model has learned noise specific to the training set. (Low scores on both would suggest high bias / underfitting.)
Confirm: plot learning curves (training vs validation score as the training set grows); check for leakage or a distribution difference between the splits; use cross-validation rather than one split.
Fixes: more or augmented data; a simpler model or fewer features; regularisation (L1/L2, dropout, weight decay); early stopping; for trees, limit depth or use bagging/ensembles. Always compare with a simple baseline, and keep the test set untouched until the end.
Original UniCircle question. Last reviewed 21 Oct 2026.