# Why 1 out of 5 AUC cross validation score is very low?

I am using xgboost model for binary classification problem. I am using 5 fold cross-validation (stratified as class imbalance) which results into the following.

xgb_model = xgb.XGBClassifier(objective="binary:logistic", random_state=42)

scores = cross_val_score(xgb_model, X, y, scoring="roc_auc", cv=5)

>> [0.97167546 0.96833211 0.58207582 0.90730687 0.93105652]


Why is one out of the above five AUC scores very low(~0.58207582)?

• I would investigate the contents of the fold. Perhaps they are harder examples, for whatever reason? Or by coincidence they're all different in some way from the training data for that fold? – Sycorax says Reinstate Monica Feb 14 at 17:37
• Is it reproducible with a different random state value (or without any)? – itdxer Feb 14 at 17:39
• @itdxer Yes, I just checked and it gives the same results. – Arch Desai Feb 14 at 17:42
• Can you please provide a reproducible example? – usεr11852 Feb 25 at 15:19