Time Robust Trees: Using Temporal Invariance to Improve Generalization

Abstract

As time passes by, the performance of real-world predictive models degrades due to distributional shifts and learned spurious correlations. Typical countermeasures, such as retraining and online learning, can be costly and challenging in production, especially when accounting for business constraints and culture. Causality-based approaches aim to identify invariant mechanisms from data, thus leading to more robust predictors at the possible expense of decreasing short-term performance. However, most such approaches scale poorly to high dimensions or require extra knowledge such as data segmentation in representative environments. In this work, we develop the Time Robust Trees, a new algorithm for inducing decision trees with an inductive bias towards learning time-invariant rules. The algorithm’s main innovation is to replace the usual information-gain split criterion (or similar) with a new criterion that examines the imbalance among classes induced by the split through time. Experiments with real data show that our approach improves long-term generalization, thus offering an exciting alternative for classification problems under distributional shift.