XLinear: A Lightweight and Accurate MLP-Based Model for Long-Term Time Series Forecasting with Exogenous Inputs

Authors: Xinyang Chen, Huidong Jin, yu huang, Zaiwen Feng Date: 2026-03-14 Paper ID: openalex:2601.09237

Summary

The paper introduces XLinear, a lightweight Multi-Layer Perceptron (MLP) based model tailored for long-term time series forecasting, specifically designed to incorporate informative exogenous variables efficiently. XLinear employs a global token, derived from the primary endogenous variable, to serve as a central hub for interacting with and weighting the influence of external inputs. The model utilizes MLPs with sigmoid activation to effectively disentangle and model both temporal autocorrelation and cross-variate dependencies introduced by exogenous features. Evaluation across seven standard and five real-world datasets demonstrates that XLinear achieves superior accuracy and computational efficiency compared to existing state-of-the-art forecasting methods when exogenous data is available.

Key Contributions

• Proposed XLinear, a lightweight MLP-based model designed to efficiently exploit informative signals from both temporal dynamics and relevant exogenous variables for long-term forecasting.
• Introduced a global token derived from the endogenous variable acting as a pivotal hub to selectively extract and integrate relevant information from associated exogenous variables.
• Achieved superior accuracy and efficiency compared to state-of-the-art models across seven standard benchmarks and five real-world datasets featuring exogenous inputs.
• Demonstrated the effectiveness of using MLPs with sigmoid activation to explicitly model both temporal patterns and variate-wise dependencies in the presence of cost-effective exogenous data.

Limitations

The paper primarily focuses on leveraging exogenous inputs and its performance against modern Transformer or State Space Model baselines without exogenous inputs for purely endogenous long-term forecasting is not explicitly detailed in the abstract.

Open Questions & Future Work

• reducing-vgm-dimensionality-in-high-dimensional-forecasting
• optimizing-lookback-windows-and-bolstering-reliability

Key Concepts

• XLinear: A lightweight, MLP-based time series forecasting model that uses a global token derived from the endogenous variable to selectively interact with exogenous variables.

Datasets

• ETTh1

Limitations

The paper primarily focuses on leveraging exogenous inputs and its performance against modern Transformer or State Space Model baselines without exogenous inputs for purely endogenous long-term forecasting is not explicitly detailed in the abstract.

Links

• Abstract
• PDF