问题
I'm trying to build a classification model and my target is not binary. The correlations of my features against my target are all weak (mostly 0.1). I have preprocessed my data and applied the all the algorithms i used to it (the algorithms i used are svm, knn, naivebayes,logistic regression, decision tree,gradient boosting, random forest). I evaluated all of the models with sklearn metrics.accuracy_score just to know how good they perform on my data but all of them scored 0.1~0.2 . The target is productline column.
My questions
- How could this happen?
- How to tackle this issue?
- Is there any other algorithm that could make better score?
回答1:
What's the accuracy if you use a dummy classifier? The accuracy of the models you have tried should be at least equal to that of the dummy classifier.
"How could this happen?" If there's no relationship between the features and the target variable, the model isn't going to return good results.
I'm not sure about the details of your dataset, but you can try to 1) Get more data 2) Get more features 3) Do some feature engineering 4) Clean your dataset if you haven't, there might be outliers or wrong inputs affecting your results
来源：https://stackoverflow.com/questions/61316262/poor-accuracy-score-on-classfication-problem