notesum.ai

Published at October 30

FlexTSF: A Universal Forecasting Model for Time Series with Variable Regularities

cs.LG

cs.AI

Released Date: October 30, 2024

Authors: Jingge Xiao¹, Yile Chen², Gao Cong², Wolfgang Nejdl¹, Simon Gottschalk¹

Aff.: ¹L3S Research Center, Leibniz Universität Hannover; ²Nanyang Technological University

Arxiv: http://arxiv.org/abs/2410.23160v1

Refer to caption

		Regular Time Series						Irregular Time Series
		ETTh	ETTm	ExRate	Illness	Weather	HAR-Phone	METR-LA	ArabDigit	CharTraj	HAR-IMU	eICU	PhysioNet12
	Lag-Llama	0.649	0.967	0.156	0.931	0.843	0.882	0.787	0.712	0.755	0.640	NaN	NaN
	MOIRAI	0.267	0.470	0.282	1.805	0.246	0.715	0.659	0.841	0.879	0.526	NaN	NaN
	TimesFM	0.344	0.741	0.127	2.729	0.539	0.948	0.741	0.803	1.150	0.857	0.996	1.143
\hdashline
	FlexTSF	0.230	0.235	0.043	2.751	0.144	0.615	0.516	0.646	0.656	0.418	0.674	0.583