Edit: Adding a repex of 10 points of data.

data = {
    "time": [327, 330, 333, 336, 339, 342, 345, 348, 351, 354],
    "resistance": [140000.0, 139000.0, 137000.0, 136000.0, 134000.0,
                   134000.0, 133000.0, 132000.0, 132000.0, 131000.0]
}
df = pd.DataFrame(data)

I am trying to fit an XGBRegressor to predict future behavior of a sensor. Data is seasonal with 6 Cycles. When I do fit the data, the RMSE for the training decreases while for Testing it doesn't decrease much and starts increasing. If I change the learning rate to 1 and max depth to 3 it overfits on training data but is a straight line on the testing. Here is the code for the model Prediction of Training data

from xgboost import XGBRegressor
model = XGBRegressor(n_estimators = 1000, early_stopping_rounds = 50,
                   learning_rate = 0.01, max_depth = 5)
model.fit(X_train, Y_train, 
        eval_set = [(X_train, Y_train), (X_test, Y_test)],
        verbose = 10)

# Make predictions on the training data
Y_train_pred = model.predict(X_train)

For reference: X has 3 features which are time (each point is 3 seconds), other 2 are

sin_time = 0.5 * np.sin(time) * 2 * np.sin(time) * time  
sin_2pi_time = np.sin(2 * np.pi * time)

Y is resistance

Tried changing the parameters but with no luck and even If I overfit the model the training data prediction is a straight line Prediction of Testing data

Here is the decision tree of 5 max depth visualized by graphviz tree

Also, here is the importance of the 3 features printed by feature_importances_

[9.9349368e-01 5.7454295e-03 7.6084811e-04]