Motif-aware Graph Neural Networks for Networked Time Series Imputation

Authors: Nourhan Ahmed, Vijaya Krishna Yalavarthi, Lars Schmidt-Thieme

AAAI 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results demonstrate that when compared to state-of-the-art models for time-series imputation tasks, our proposed model can reduce the error by around 19%. Experiment. Empirical results show that Motif-GNN outperforms baseline models by approximately 19% for time series imputation across various real-world datasets. Evaluation is conducted using Mean Absolute Error (MAE) and Mean Relative Error (MRE) as metrics.
Researcher Affiliation Collaboration 1 Information Systems and Machine Learning Lab (ISMLL), University of Hildesheim, Germany 2 VWFS Data Analytics Research Center EMAIL
Pseudocode Yes Algorithm 1: The Framework of Motif-GNN.
Open Source Code No No information about open-source code is provided in the paper. There are no links to repositories or explicit statements about code release.
Open Datasets Yes The air quality dataset comprises observations from 36 monitoring sites (Yi et al. 2016), collected between May 1, 2014, and April 30, 2015, totaling 8,759 timestamps (Yi et al. 2016; Zheng et al. 2015). Traffic Datasets. The traffic datasets used in this paper are Pe MS-LA, Pe MS-BA, and Pe MS-SD, collected from January 1 to March 30, 2022, from highways in Los Angeles County, the Bay Area, and San Diego, respectively (Chen et al. 2001).
Dataset Splits Yes Under the MCAR scenario, datasets are split into 70% for training, 10% for validation, and 20% for testing. In the block-missing scenario, we simulate overlapping time- and space-structured (spatio-temporal) missingness by randomly dropping 5% of the data per sensor and simulating failures with a 0.15% probability, varying the duration from 1 to 4 hours for traffic data and 2 to 6 days for air quality data, following the evaluation protocol in (Cini, Marisca, and Alippi 2022). Additionally, to ensure consistency with baseline models, we consider time-structured missingness in the air quality dataset by partitioning the 1-year data into two parts, using the months of March, June, September, and December as the test set and the rest for training (Yi et al. 2016; Cini, Marisca, and Alippi 2022; Wang et al. 2023a). For traffic datasets, we used the first half of February and the last half of March for testing and the rest for training.
Hardware Specification No The paper does not provide specific hardware details such as GPU models, CPU types, or memory amounts used for running the experiments.
Software Dependencies No The paper does not provide specific software dependency details with version numbers, such as programming language versions or library versions (e.g., Python 3.x, PyTorch 1.x).
Experiment Setup No The paper does not provide specific experimental setup details such as concrete hyperparameter values (e.g., learning rate, batch size, number of epochs) or optimizer settings in the main text.