In the previous section we investigated the impact of datasets’ meta-features on the results of benchmarking and modeled the difference between methods’ performance based on these meta-features. In this section, we take a different approach for the explanation of differences. We use partial dependence plots as a technique to assess the dependency pattern between response and features underlying the prediction rule. More precisely, the aim of these additional analyses is to assess whether differences in performances (between LR and RF) are related to differences in partial dependence plots. After getting a global picture for all datasets included in our study, we inspect three interesting “extreme cases” more closely. In a nutshell, we observe no strong correlation between the difference in performances and the difference in partial dependences over the 243 considered datasets. More details are given in Additional file 3: in particular, we see in the third example dataset that, as expected from the theory, RF performs better than LR in the presence of a non-linear dependence pattern between features and response.
