Hierarchical Time Series Forecasting Via Latent Mean Encoding

2025-06-24

AI Summary

This paper addresses the challenge in temporal hierarchical forecasting where coarse- and fine-grained target variables exhibit inconsistent behavioral patterns, hindering profit-driven decision-making. We propose a novel hierarchical time-series forecasting framework that models the global mean pattern of the target variable via latent encoding and employs dedicated prediction modules for each temporal aggregation level to ensure cross-granularity consistency and synergy. Its core innovation is a latent mean encoding mechanism that explicitly enforces statistical consistency across hierarchy levels, thereby enhancing overall predictive reliability. Experiments on the real-world M5 dataset demonstrate that our method significantly outperforms state-of-the-art baselines—including TSMixer—achieving superior performance in both MAE and RMSE. These results validate its effectiveness in supporting fine-grained operational decisions in commercial applications.

Technology Category

• Machine Learning: Time-Series/Data Streams
• Planning, Routing, and Scheduling: Temporal Planning
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Algorithms and analysis for heterogeneous, signed, attributed, multi-relational, temporal, higher-order, and annotated Web-related graphs
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Coherently forecasting the behaviour of a target variable across both coarse and fine temporal scales is crucial for profit-optimized decision-making in several business applications, and remains an open research problem in temporal hierarchical forecasting. Here, we propose a new hierarchical architecture that tackles this problem by leveraging modules that specialize in forecasting the different temporal aggregation levels of interest. The architecture, which learns to encode the average behaviour of the target variable within its hidden layers, makes accurate and coherent forecasts across the target temporal hierarchies. We validate our architecture on the challenging, real-world M5 dataset and show that it outperforms established methods, such as the TSMixer model.

Problem

• Forecasting target variable across multiple temporal scales
• Ensuring coherent predictions in hierarchical time series
• Improving accuracy in profit-optimized business decision-making

Innovation

• Hierarchical architecture for multi-scale forecasting
• Latent mean encoding for coherent predictions
• Specialized modules for temporal aggregation levels