I am dealing with a house prediction problem. However, it has about 10% missing values in buildYear which is one of the most important features. I tried filling them with mean value or mode value. Of course, it gets increased in local cv and online validation like a leaderboard in kaggle.

One day, I got inspired by others to use a model's prediction to fill its missing values. So, I use Logistic Regression's predictions to fill the missing values in training set and use the target variable: tradeMoney

Then, I filled missing values in the test set casually with its mode values (I know it is a terrible way). I then got a higher value in the local CV and online validation.

So, I am wondering, maybe I can choose other features to fill its missing values. I can't use brute force with every possibility for submission times limits.

BUT I don't know how to choose these features as training values to get a higher prediction performance and avoid overfit and what conditions we should consider like correlations or something else. And its highest correlations with other feature is about 0.4.

Can anyone give me some hints or references?


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.