Technology Category
Machine Learning: Time-Series/Data StreamsPlanning, Routing, and Scheduling: Planning with Language ModelsNatural Language Processing: (Large) Language Models
Benchmarking Time Series Forecasting Models: From Statistical Techniques to Foundation Models in Real-World Applications
🤖 AI Summary
This work addresses large-scale, distributed forecasting in the hospitality industry by establishing a benchmark for 14-day hourly sales prediction across thousands of restaurants in Germany, incorporating multi-source features including weather, calendar events, and temporal patterns.
Method: We first empirically validate the zero-shot generalization capability of foundational time-series models—Chronos and TimesFM—on real-world industrial data; propose a hybrid forecasting paradigm integrating ML-based meta-models with foundational models; and implement an efficient, scalable computation framework leveraging PySpark-Pandas interoperability.
Contribution/Results: (1) Foundational models achieve accuracy comparable to optimized XGBoost/LightGBM—without manual feature engineering—demonstrating their practical viability for industrial forecasting. (2) The proposed architecture supports horizontal scaling to thousands of outlets, reduces end-to-end inference latency by 40%, and significantly enhances the deployability and operational utility of large-scale time-series forecasting systems.
Technology Category
Application Category
📝 Abstract
Time series forecasting is essential for operational intelligence in the hospitality industry, and particularly challenging in large-scale, distributed systems. This study evaluates the performance of statistical, machine learning (ML), deep learning, and foundation models in forecasting hourly sales over a 14-day horizon using real-world data from a network of thousands of restaurants across Germany. The forecasting solution includes features such as weather conditions, calendar events, and time-of-day patterns. Results demonstrate the strong performance of ML-based meta-models and highlight the emerging potential of foundation models like Chronos and TimesFM, which deliver competitive performance with minimal feature engineering, leveraging only the pre-trained model (zero-shot inference). Additionally, a hybrid PySpark-Pandas approach proves to be a robust solution for achieving horizontal scalability in large-scale deployments.
Problem
Research questions and friction points this paper is trying to address.
Evaluates forecasting models for sales
Compares statistical, ML, and foundation models
Focuses on scalability in distributed systems
Innovation
Methods, ideas, or system contributions that make the work stand out.
ML-based meta-models outperform others
Foundation models require minimal feature engineering
Hybrid PySpark-Pandas ensures scalability
🔎 Similar Papers
No similar papers found.
