Temporal Causal Discovery and Generative Prediction of Vehicular CO$_2$ emission

ICLR 2025 Conference Submission13556 Authors

Keywords: causal discovery, causal inference, time series forecasting, vehicular emission

TL;DR: Temporal causal discovery and causal forecasting for vehicular carbon dioxide emission prediction

Abstract: Global warming from greenhouse gas emissions is humanity's largest environmental hazard. Greenhouse gases, like CO$_2$ emissions from transportation, notably cars, contribute to the greenhouse effect. Effective CO$_2$ emission monitoring is needed to regulate vehicle emissions. Few studies have predicted automobile CO$_2$ emissions using OBD port data. For precise and effective prediction, the system must capture the underlying cause-effect structure between vehicular parameters that may contribute to the emission of CO$_2$ in the transportation sector. Thus, we present a causal RNN-based generative deep learning architecture that predicts vehicle CO$_2$ emissions using OBD-II data while keeping the underlying causal structure. Most widely used real-life datasets lack causal relationships between features or components, so we use our proposed architecture to discover and learn the underlying causal structure as an adjacency matrix during training and employ that during forecasting. Our framework learns a sparse adjacency matrix by imposing a sparsity-encouraging penalty on model weights and allowing some weights to be zero. This matrix is capable of capturing the causal relationships between all variable pairs. In this work, we first train the model with widely used synthetic datasets with known causal structure among variables, then we apply it to the state-of-the-art OBD-II dataset to find the internal causal structure among the vehicular parameters and perform causal inference to predict CO$_2$ emission. Experimental results reveal that our causal discovery and forecasting method surpasses state-of-the-art methods for the tasks of causal discovery in terms of AUROC, forecasting on multivariate causal time series data, and OBD-II dataset in terms of MMD, RMSE, and MAE. After successful completion, we will release the code (Code for review - \href{https://anonymous.4open.science/r/causal-obd-co2-0A0C}{https://anonymous.4open.science/r/causal-obd-co2-0A0C}).

Primary Area: causal reasoning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 13556