XGBoost Classification with highly unbalanced dataset - classification

Question

I am trying to make a classification with XGBoost (using the Scikit-Learn API).
The problem is that, when the dataset is highly unbalanced, using the parameter scale_pos_weight does not help, because the resulting balanced accuracy is very low (0.5).
I think the scale_pos_weight parameter is not taken into account when using the eval_set.
Can you help? This is an example code:
from sklearn.model_selection import train_test_split
from sklearn.datasets import make_classification
from sklearn.preprocessing import StandardScaler
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score
from sklearn.metrics import balanced_accuracy_score
import xgboost as xgb
# create huge dataset
X, y = make_classification(n_samples=100000,
n_features=10,
n_informative=10,
n_redundant=0,
n_classes=2,
weights=[0.999],
random_state=1)
# split into train/test sets with same class ratio (stratify=y)
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33, stratify=y, random_state=1)
# standardize the data
scaler = StandardScaler()
X_train = scaler.fit_transform(X_train)
X_test = scaler.transform(X_test)
# train the classifier
model = xgb.XGBClassifier(n_estimators=1000, scale_pos_weight=1/999)
model.fit(X_train, y_train,
early_stopping_rounds=100,
eval_metric='logloss',
eval_set=[(X_test, y_test)],
verbose=False)
# best iteration
print('best iteration', model.best_iteration)
best_iteration = model.best_ntree_limit#clg.get_booster().best_ntree_limit
#print(model.evals_result())
y_pred = model.predict(X_test, ntree_limit=best_iteration)
#y_pred = model.predict(X_test)
print('accuracy', accuracy_score(y_test, y_pred))
print('balanced_accuracy_score', balanced_accuracy_score(y_test, y_pred))