I have gotten a dataset that has continuous values which are normalized in the range from 0 to 1, is it approximately 200 records with 14 features, and like 30 percent of the data present there are outliers. I have tried to use the Isolation model of Scikit, and now I am performing the hypeparameters tuning. So far what I have done is the following:

parameters = {'n_estimators':[10,20,40,60,80,100], 'max_features':[0.1, 0.2, 0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0], 'max_samples':[10,20,30],'bootstrap':[True,False]}
    clf = GridSearchCV(iso, parameters,  scoring=scoring,n_jobs=-1,verbose=5,cv=10)

and for the scoring function I have used a weighter F1-score. The problem is when I get the best hyperparameters results, for example:

best parameters  {'bootstrap': False, 'max_features': 0.3, 'max_samples': 10, 'n_estimators': 40}

and I apply these values I got some erratic behaviour everytime that I run my model to obtain the ROC curve. For example in one run I can get:

but when I run again my model with the same hyperparameters I can get something like this:

enter image description here

why this could be happening?



Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.