Depending on the modeling technique that you're using, some of this work may be more valuable than other parts. Deep learning algorithms tend to perform better on less-engineered data than shallower models and it might be that less work is needed to improve results.
The key to understanding what is needed is to iterate quickly through the whole process from dataset acquisition to modeling. On a first pass with a clear target for model accuracy, find the acceptable minimum amount of processing and perform that. Learn whatever you can about the results and make a plan for the next iteration.
To show how this looks in practice, we'll work with an unfamiliar, high-dimensional dataset, using an iterative process to generate increasingly effective modeling.
I was recently living in Vancouver. While it has many positive qualities, one of the worst things about living in the city was the somewhat unpredictable commute. Whether I was traveling by car or taking Translink...