dependency analysis using copulas
Budget: $30 – $250 USD
I have a copulas model built that I need to investigate further. All the data is already collected and sorted. I can provide html files, so you can see more information of what's already done. But there is further analysis listed in the deliverables section needs to be completed. I need somebody who's good with copulas modelling and specifically is able to meet the deliverables.
Deliverables
• Do the Hoeffding’s D independence test for the example copulas models.
• More analysis of independent wind farms (project lead was very happy about this part but wanted more) – so similar to the outliner analysis that done already, need to do more analysis. (where are the wind farms which are, like completely unrelated to other wind farms (have very independent profiles seen in the Pearson’s correlation and copulas), identify them and explain why. Not necessarily the outlier’s but the ones that have low dependencies when compared to other wind farms on that graph.
• Clusters - Looking at the clusters that are in both the Pearson’s correlation and copulas, make another Pearson's correlation and copulas of these clusters separately and compare against the main one. (so, comparing to main Correlation vs Copulas and see how they differ). One cluster with very good dependency, two is medium level dependency and 3 is no dependency.
• Regress the output of one wind farm against another one in a linear regression model, then test how well does that fit the data as compared to the copulas model. That will be interesting because then you actually measure whether the copulas does a better job (i.e does this nonlinear relationship that you captured in the copulas model actually improve (or generally is a better model) than the assuming a linear relationship which is what the Pearson's correlation does. You take a pairs of wind farms, I’m going to get wind farm A, Wind farm A is the independent variable and wind farm B is the dependent variable, regress them each other with no other variables, analysis the r^2, come with simple metric for the copulas and compare whether it is more accurate.
Deliverables
• Do the Hoeffding’s D independence test for the example copulas models.
• More analysis of independent wind farms (project lead was very happy about this part but wanted more) – so similar to the outliner analysis that done already, need to do more analysis. (where are the wind farms which are, like completely unrelated to other wind farms (have very independent profiles seen in the Pearson’s correlation and copulas), identify them and explain why. Not necessarily the outlier’s but the ones that have low dependencies when compared to other wind farms on that graph.
• Clusters - Looking at the clusters that are in both the Pearson’s correlation and copulas, make another Pearson's correlation and copulas of these clusters separately and compare against the main one. (so, comparing to main Correlation vs Copulas and see how they differ). One cluster with very good dependency, two is medium level dependency and 3 is no dependency.
• Regress the output of one wind farm against another one in a linear regression model, then test how well does that fit the data as compared to the copulas model. That will be interesting because then you actually measure whether the copulas does a better job (i.e does this nonlinear relationship that you captured in the copulas model actually improve (or generally is a better model) than the assuming a linear relationship which is what the Pearson's correlation does. You take a pairs of wind farms, I’m going to get wind farm A, Wind farm A is the independent variable and wind farm B is the dependent variable, regress them each other with no other variables, analysis the r^2, come with simple metric for the copulas and compare whether it is more accurate.