CODEWITHSUNDEEP
machine-learning
user12345
user12345

Reputation: 29

Interdependece of parameters in Machine Learning problems

Let's say we have 10-D data from class of students. The data involves parameters like Name, Grades, Courses, No. of hours of lectures, etc. of all the students of the class. Now, we want to analyze the impact of No. of hours of lectures on Grades. If we closely watch our parameters, Name of the student has nothing to do with Grades, but Courses taken by student "might" have impact on Grades.

So, there could be parameters which are dependent on each other while some others can be totally independent. My question is, how do we decide that which parameter has impact on our classification/regression problem and which don't?

PS: I am not looking for exact solutions. If someone can just show me the right direction or keywords for google search, that should be sufficient. Thanks.

Upvotes: 1

Views: 155

Answers (2)

Don Reba
Don Reba

Reputation: 14031

This is the problem of independent component analysis. ICA a family of methods for finding the statistically independent components of data sets. This is a difficult problem, and there exists a large variety of algorithms for finding good solutions. A popular algorithm is FastICA.

There are also related concepts of whitening and decorrelation.

Upvotes: 1

zinglon
zinglon

Reputation: 1908

The technique you're looking for is called dimension reduction. The Stanford machine learning class goes over one method (principal component analysis).

Upvotes: 1

Related Questions