Global Study: Influence of Tuning

AbstractExpanding the more focused analyses from previous chapters, this chapter takes a broader view at the tuning process. That means, rather than tuning an individual model, this investigation considers the tuning of multiple models, with different tuners, and varying data sets. The core aim is to see how characteristics of the data and the model choice may impact the tuning procedure. We investigate five hypotheses, concerning the necessity of tuning, the impact of data characteristics, the impact of the target variable type, the impact of model choice, and benchmarking. Not only does this entail an in-depth tuning study, but we also tie our results to a measure of problem difficulty and use consensus ranking to aggregate the diverse experimental results.