О
May the force be with you
Size: a a a
2019 July 22
LS
Haha thanks
LS
I am mixed dataset. With 150 features (96 numeric /rest categorical) what is the best way to do feature selection before applying any ML algorithm?
LS
О
Why do you want to reduce features?
LS
Because performance is very bad with all feature
LS
I thought it may help for better performance
О
Have you tried PCA?
DP
I am mixed dataset. With 150 features (96 numeric /rest categorical) what is the best way to do feature selection before applying any ML algorithm?
are all of this 96 numerical features indepentant of each other?
LS
I tried PCA after 1 hot encoding of categorical. But not suitable as Principal component 1 only explains 6% variance
LS
are all of this 96 numerical features indepentant of each other?
How to find out if independent?
DP
How to find out if independent?
you were mentioning 1 hot encoding of age
so that you have ~60 inputs for age (from 20 to 80 years), right?
are this 60 input for age included in this 96 features as 1 feature or as 60?
so that you have ~60 inputs for age (from 20 to 80 years), right?
are this 60 input for age included in this 96 features as 1 feature or as 60?
LS
No age is only 1 feature /column like 18,30,25
LS
I mean 1 hot encoding of categorical features not numeric
DP
okay, thats good
DP
is this task solvable by a human with the performance you want?
how do people detect cancer in your data?
how do people detect cancer in your data?
LS
There is a target variable which says : healthy /cancer
DP
There is a target variable which says : healthy /cancer
can people guess it from the data available to your forest?
LS
No it's not possible
LS
There are 1800 rows and 150 features