-1
$\begingroup$

I want your help to analyze this correlation heat map
If you look at this heat map, what can you infer and what can you apply?

enter image description here

Improve this question
$\endgroup$

3 Answers 3

Reset to default
1
$\begingroup$

From this heatmap, you could infer the correlation between features, if they are positively correlated, negatively correlated, or not correlated at all. The intensity of the colors shows the correlation. Dark red, strongly positively correlated, i.e., if the value in one of the features increases, the value of the other feature increases as well. Dark blue indicates a negative correlation, i.e., if a value of one feature increases, the value of the other feature decreases. White refers to no correlation.

Hope that helps.

Improve this answer
$\endgroup$
2
  • 1
    $\begingroup$ thx if you apply this to the Randomforest,lgb or xgboost what whould you do? you would find any features to remove or something else? $\endgroup$
    – slowmonk
    Commented Feb 14, 2020 at 11:22
  • 1
    $\begingroup$ @slowmonk based on the problem you have at hand, you decide this. Ideally when you do feature reduction, you end up removing highly correlated features. Do take a look at PCA and tsne for feature reduction. You would evalute your model with certain domain knowledge to drop certain features as a permutation combination approach. This could help your model to predict with better accuracy. $\endgroup$
    – Nischal Hp
    Commented Feb 14, 2020 at 11:30
0
$\begingroup$

This sounds like a feature engineering exercise. Check the links below for relevant info.

Feature Selection Techniques in Machine Learning with Python | Towards Data Science

A Feature Selection Tool for Machine Learning in Python | Towards Data Science

And this last one, directly below, is probably the easiest to implement, and probably the best case for your scenario.

http://blog.yhat.com/tutorials/5-Feature-Engineering.html

Improve this answer
$\endgroup$
1
  • $\begingroup$ The last link is not working. $\endgroup$
    – Shayan Shafiq
    Commented Mar 30, 2021 at 14:22
0
$\begingroup$

Looks like your features psfMag_u and fiberMag_u are highly correlated. Remove one of them. Also, psfMag_u and petroMag_u are negatively correlated. So you can get rid of one of them.

Other than that, nothing much can be deduced just from this graph.

Improve this answer
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.