I created a little model that predicts the time a task will take to execute. Than, I calculate the real time that the task took to execute. With the predict and real value I have the error of my prediction. I want to use that error to predict the next execution (the next execution is a similar task has the one I ran). I was looking to R-squared or coefficients to help me. Does anyone has an opinion about this. What is the best use I can do to the error of my prediction to help to predict again? For now I am using Adjusted R^2, but I don't know other methods.