How to make predictions based on correlations?

I have correlation values for profit based on three different attributes - attribute1,attribute2,attribute3. Now I want to make predictions for the profit based on these 3 different correlation values.

I wanted suggestions for some efficient algorithms that can help me to make predictions from these correlation values.

Note: I am new to this so I would appreciate some suggestions if there is some lack of understanding in me.

2 Answers

The correlation coefficient will tell you, how much predictive power the individual attributes posess with respect to profit in a linear sense. Easier: The correlation coefficient tells you how line-like your data is. ( There might still be a very predictive arc like structure in your data which the correlation coefficient can't capture )

So you can either calculate a regression line for profit with the attribute having the highest correlation coefficient, or you can calculate a hyperplane for all three attributes to get the desired prediction. Regression lines and hyperplanes are usually calculated by least-squares fitting.

But beware to overestimate predictive power, because it is extrapolation what you do. New data might behave different than old data which makes extrapolation potentially less exact the further you are away from known data.

I am sorry to disappoint you but Correlation does not imply causation !.

You need to make sure that the data is statically significant (enough data, the difference is not negligible, etc), I would suggest Z-Test, or chi-square, depending on the nature of your data.

• statistical significance does not imply causation either... but it doesn't appear that the user wants to make claims about the relationships rather predict. Tobias's answer hits the nail on the head. – Lauren Goodwin Sep 23 '15 at 16:46