Explore the Time Series Forecasting Potential of TabPFN Leveraging the Intrinsic Periodicity of Data

Sibo Cai, Xi Sun, Hui Zhong

Keywords: TabPFN, Time Series Forecasting, Intrinsic Data Periodicity

TL;DR: This paper proposes a TabPFN-based time series forecasting method leveraging intrinsic data periodicity. Experimental results show that our method outperforms TabPFN-TS, validating the value of integrating periodicity into TabPFN-based predictions.

Abstract: Time series forecasting has extensive and significant applications in various fields such as transportation, energy, finance, etc. In recent years, time series foundation models have emerged prominently in the prediction field due to their ability to achieve accurate predictions with minimal fine-tuning or even in zero-shot scenarios. Among these models, TabPFN and its derivative TabPFN-TS stand out as notable examples. This paper introduces a TabPFN-based time series prediction method that capitalizes on the intrinsic periodicity of data. Experimental evaluations across multiple time series datasets reveal that our proposed method outperforms TabPFN-TS. This outcome validates the efficacy of incorporating data periodicity into TabPFN-based time series prediction. The code of our method can be found here:

Submission Number: 25