Denoising matrix factorization for high-dimensional time series forecasting

The matrix factorization method (MF) has gained widespread popularity in recent years as an effective technique for handling high-dimensional time series data. By converting large-scale data sets into low-rank representations, MF-based methods have proven to be successful. However, these methods continue to face challenges in managing long-term dependencies, primarily due to the presence of noise and a lack of prior knowledge regarding the underlying matrix. To overcome this issue, we propose a novel approach that incorporates a latent bias effect and a denoising model, which enables the model to recover the underlying matrix more effectively and improves the precision of the model. By focusing only on relevant components, our proposed model constructs the underlying matrix more precisely through denoising operations. Our experiments conducted on four benchmark datasets demonstrate that our proposed model outperforms existing methods in terms of accuracy and robustness.

Automated summary

Matrix factorization method has gained popularity in handling high-dimensional time series data. However, challenges still exist in long-term dependency management. To address this, we propose a novel approach that incorporates a latent bias effect and denoising model, improving the accuracy and robustness of the model.