Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces a model-agnostic methodology for measuring lag relevance in univariate time series forecasting, utilizing Ghost variables and Shapley values alongside additive importance measures. The authors propose the auto-relevance and partial auto-relevance functions as new metrics for assessing lag importance, and they present a method to substitute missing features with one-step forecasts from the same model. Evaluations across various simulations and real datasets reveal that these relevance measures effectively capture the expected lag structure in nearly all scenarios tested.
The new auto-relevance functions reveal lag importance in time series forecasting with unprecedented clarity, outperforming traditional methods.
We propose a model agnostic methodology to measure lag relevance in machine learning forecasting models applied to univariate time series. Particularly, we are working in the context of time series using the frameworks of Ghost variables and Shapley values, together with additive importance measures, to introduce the auto-relevance and partial auto-relevance functions as the lag importance values. Additionally, we propose a novel method to replace absent features in coalition based methods with a one step forecast from the same model. We evaluate these proposals under different simulations and real data cases. This combined framework perspective is particularly suitable for time series. In addition, to show our discoveries we use a pull of models from the seasonal ARMA family and recurrent neural networks. We found that the calculated relevance measures successfully demonstrate the expected lag structure in almost all cases.