With logistic regression results, we can explain the impact of each feature by using regression coefficients, and identify big influencers by comparing those coefficients.

With the same logic, we can also rank each feature by its effects as calculated by the logistic regression coefficients.

Another way is to use the R package of effect, which was created by John Fox and others especially for the display of the effects of linear and generalized linear models. By using this package, we can obtain a list and some graphical displays with the function of plot (effect) .

To take this to a higher level, users may consider using the R package relaimpo, specially created to assess the relative importance of predictors, for which the following code should be used:

library(relaimpo)
calc.relimp(model1, type = c("lmg"), rela = TRUE)

Here, "lmg" refers to the authors Lindeman, Merenda, and Gold for their special methods.

As used in the previous chapters, with the randomForest package in R, a simple code of estimatedModel$importance will return a ranking of variables by order of their importance in determining the default risk.

However, to obtain the importance of variables through randomForest functions, we need a full model estimated, with all data complete. So it does not really solve our problems.

For this project, we will rely on results obtained from logistic regression.

As for our client, we find a few very interesting insights, among which are a few features with big impacts, which include the length of stay at the current address and social media influence.