Mission accomplished
The mission was to understand how fuel efficiency was impacted over the years by the potential predictors in the dataset. We determined that the most significant fuel efficiency predictors, by far, are pollution-related, and that tailpipe CO2 in grams/mile (co2TailpipeGpm) is the one that stands out. Both pollution and fuel inefficiency decrease with every year. Likewise, they increase with the number of cylinders and when it's a diesel engine (fuelType_Diesel). None of this should be surprising to anybody who knows about cars' evolution over the past few decades.
However, there were some revealing insights. For instance, SHAP dependence plots (Figures 5.12 and 5.14) helped us understand why the co2 and ghgScore features are redundant. And as depicted by an interaction ALE plot (Figure 5.19) there might be some data quality issues with co2TailpipeGpm before 2004, which should be investigated further. Global surrogates distilled a sense of hierarchy...
