Mixed logit models with unobserved inter- and intra-individual heterogeneity were tested for their ability to predict individual choice behavior accurately. The study found that these models do not significantly improve choice prediction compared to standard mixed logit models. Even with high levels of taste variation within individuals and many choice situations, the more complex models did not show clear benefits. Additionally, the more advanced models required significantly more computation time for estimation. Alternative modeling approaches that can capture richer dependencies between decision-makers, alternatives, and attributes were suggested based on recent advancements in machine learning and econometrics.