CRAFT: Time Series Forecasting with Cross-Future Behavior Awareness

TL;DR

This work tackles time series forecasting under uncertainty by introducing Cross-Future Behavior ($CFB$), features that occur before the current time but affect the future. It presents CRAFT, a three-module framework consisting of the Koopman Predictor Module ($KPM$), Internal Trend Mining Module ($ITM$), and External Trend Guide Module ($ETG$) to transfer $CFB$ trends to the target series, aided by hierarchical sampling and a demand-constrained loss. A suite of losses, including $\,\mathcal{L}_{be\_k}$, $\mathcal{L}_{be\_y}$, and $\mathcal{L}_{recon}$, calibrates predictions and enforces hierarchical consistency. Experiments on a real-world, large-scale hotel booking dataset and online A/B tests demonstrate substantial improvements over strong baselines across multiple forecast horizons, with practical deployment evidenced by real-world metrics like $IWR$ and $PHDI$. The work offers a new paradigm for integrating future-aware features into TSF and suggests broad potential for applying $CFB$ in diverse domains.

Abstract

The past decades witness the significant advancements in time series forecasting (TSF) across various real-world domains, including e-commerce and disease spread prediction. However, TSF is usually constrained by the uncertainty dilemma of predicting future data with limited past observations. To settle this question, we explore the use of Cross-Future Behavior (CFB) in TSF, which occurs before the current time but takes effect in the future. We leverage CFB features and propose the CRoss-Future Behavior Awareness based Time Series Forecasting method (CRAFT). The core idea of CRAFT is to utilize the trend of cross-future behavior to mine the trend of time series data to be predicted. Specifically, to settle the sparse and partial flaws of cross-future behavior, CRAFT employs the Koopman Predictor Module to extract the key trend and the Internal Trend Mining Module to supplement the unknown area of the cross-future behavior matrix. Then, we introduce the External Trend Guide Module with a hierarchical structure to acquire more representative trends from higher levels. Finally, we apply the demand-constrained loss to calibrate the distribution deviation of prediction results. We conduct experiments on real-world dataset. Experiments on both offline large-scale dataset and online A/B test demonstrate the effectiveness of CRAFT. Our dataset and code is available at https://github.com/CRAFTinTSF/CRAFT.

Figures (14)

• Figure 1: Illustration of Cross-Future Behavior (CFB).
• Figure 2: The overview of the proposed Cross-Future Behavior Awareness based Time Series Forecasting method (CRAFT). The left decomposition part is based on DLinear. CRAFT is composed of three main parts, the Koopman Predictor Module (KPM), the Internal Trend Mining Module (ITM), and the External Trend Guide Module (ETG). KPM is used to extract the key trends of the trend of label and CFB, predicting the label in the prediction window. ITM is used to supplement the unknown area of the CFB. ETG is used to acquire more representative trends from higher levels.
• Figure 3: Demand-Constrained Loss.
• Figure 4: Case study on (a)-(d) four different cases, where blue line is the truth value and the orange line is the prediction of CRAFT.
• Figure 5: Cross-Future Behavior (CFB) on hotels booking example.