Poor performance of isolation forest for anomaly detection

Question

I have trained sklearn Isolation Forest on an anomalous dataset (90% normal and 10% anomaly included) with different number of estimators and different number of sample fraction