I am trying to build a regression model to predict Gerrit code review delay (i.e the time between the creation time of the code review until the time of the last update.) For that, I used a random forest regression to predict the outcome variable based on metrics describing the code review process, code complexity and developers collaboration, etc. The model is performing well on training data achieving 0.91 as an R-square score, however, on test data, I only get 0.09. When analyzing residuals for my model they don't seem to be normally distributed I think residuals have a trend of multiple increasing lines that I could not explain. (see the attached residual plot for test data
)
PS: I used log-transformed variables: Xs and y are on the log scale.
