Continuous Multi-step Predictions of Highly Imbalanced Multivariate Time Series via Deep Learning Network
Primary Area: general machine learning (i.e., none of the above)
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.
Keywords: highly imbalanced data, multivariate time series, LTV study, feature learning
Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.
Abstract: Multi-step prediction of multivariate time series has always been a very popular research topic across industries. We focus on the scenario in which the data with severe imbalance problem caused by the “0” expansion in regression analysis, and meanwhile the data contains complex textual information. Such data is very common in customer's life time value evaluation tasks in businesses. The commonly used two-stage modeling scheme effectively predicts whether or not a customer will pay for a product or service at the next moment. However, it is incapable of continuously forecasting potential payment values due to the strong imbalanced and randomness distribution of the data. In this paper, we propose a feature learning based deep learning method for imbalanced multivariate time series (FLIMTS). The innovative use of a weighted quantile loss in our proposed method handles the highly imbalance problem in regression. Furthermore, FLIMTS incorporates both the customer's payment sequence and the behavioral characteristics of their interests which allows for more accurate predictions. Empirical analysis shows that FLIMTS has significant advantages and performs better than the existing two-stage approaches on common model evaluation criteria.
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: 149
Loading