Linear Regression

[maw-cu]

I have managed to implement linear regression in the activity browser, with the calculations done by scikit-learn, if anyone is interested.

[marc-vdm]

This sounds interesting! Could you explain a bit more about what you did, what you can do with the regression tool and how it works (both technically and for users)?

[maw-cu]

Well, sensitivity analysis methods can be classified as a) regression-based, b) variance-based, c) derivative-based, etc.....

It was while I was upgrading the sensitivity analysis section to include more methods that I thought of introducing linear regression to the activity browser. I thought that this would be a useful separate LCA result subtab (next to sensitivity analysis) for assessing less complex models with limited parameter interactions.

For now, I can express the model output, which can be CO2 emissions, as a function of input parameters through a linear relation. The final goal is to compute standardized regression coefficients (SRCs), which can function as delta or s-index coefficients to also rank input parameters according to their importance and influence on model output.

[marc-vdm]

Exciting stuff, though I must admit it's not entirely clear to me what the regression coefficients would be? In case of using ecoinvent, would you have ~20k regression variables? Could you perhaps refer to papers going further into the method?

[bsteubing]

@maw-cu This is a great initiative! Would you be willing to explain what you did in a personal meeting? Please contact us on our Leiden email accounts for that. Then we could also discuss the possibility of making your features available within the AB.