Only Two Can Tango at the Pareto Front

This study examines the impact of adding dimensions (objective functions) to the optimization process and explores how prioritizing certain metrics within Pareto front calculations affects model quality in Marketing-Mix Modelling (MMM). Using Meta’s Robyn library and Nevergrad for optimization, we evaluated models across NRMSE, Decomp.RSSD, KL Divergence, and MAPE. Results indicate that, as anticipated, augmenting the number of optimization dimensions leads to a deterioration in model performance, specifically in terms of metric optimality, which in turn results in less reliable insights.