gradient Boosting Missing values

k_vishnu772

Hi All,

 

I am using a small data for binary classification problem with  320 rows. so as it is a small data set i am using all the data for training using cross validation.I have some missing values in the data and with out imputing any missnig values used gradient boosting algorithm and i got an accuracy of 83 % i would like to know what sort of missing values inputating is impeleted by gradient boosting and after that i used KNN algrrithmn to imput missing values and then applied gradint boosting and i got accuracy of 88 % .Could you please expain which one i should take as best?

 

Regards,

Vishnu

Maerkli

 

Hallo Vishnu,

 

1) As you use two different methodologies, you will get two different results.

2) You explain that have missing values in your data; do you know the percentage of the missing values? Are the missing values equally distributed on all variables? Is your data normalized before training? Did you try with different parameters?

It is perhaps not the answer you wish but I hope that it helps you to better understand your problem.

 

Maerkli

 

k_vishnu772

HI @Maerkli,

 

I have 2 features where it 30 percent of data is missing and i did't do any normalisation.i did't understand what do you mean by evenly distributed and what shoud be done if it is evenly distributed.if you could explain how should i deal this that would be very helpful for me.

 

Thank you.

Maerkli

Hi Vishnu,

 

If it is not confidential, try to share with us your XML file and your data, in .csv format, preferably.

Maerkli.