Location
Online
Event Website
https://hicss.hawaii.edu/
Start Date
3-1-2023 12:00 AM
End Date
7-1-2023 12:00 AM
Description
We introduce a general time-series forecasting method that extends classical seasonal autoregressive models to incorporate exogenous and relational information in an online setting. Our approach is implemented using the probabilistic programming language Probabilistic Soft Logic (PSL). We leverage recent work that enables the scalable application of PSL to online problems and propose novel modeling patterns to leverage dependencies between multiple time series. We demonstrate the applicability and performance of our method for the task of station-level demand forecasting on three bike sharing systems. We perform an analysis of the demand time series and present evidence of relational dependencies among the stations, motivating the need for a forecasting model that leverages the rich relational structure in the bike sharing networks. Our approach significantly improves multi-step forecasting accuracy of autoregressive time-series models on all three datasets. Further, our approach is easily extendable and we expect applicable to a variety of other time-series forecasting problems.
Recommended Citation
Dickens, Charles; Miller, Alexander; and Getoor, Lise, "Online Collective Demand Forecasting for Bike Sharing Services" (2023). Hawaii International Conference on System Sciences 2023 (HICSS-56). 3.
https://aisel.aisnet.org/hicss-56/da/decision_support_for_scm/3
Online Collective Demand Forecasting for Bike Sharing Services
Online
We introduce a general time-series forecasting method that extends classical seasonal autoregressive models to incorporate exogenous and relational information in an online setting. Our approach is implemented using the probabilistic programming language Probabilistic Soft Logic (PSL). We leverage recent work that enables the scalable application of PSL to online problems and propose novel modeling patterns to leverage dependencies between multiple time series. We demonstrate the applicability and performance of our method for the task of station-level demand forecasting on three bike sharing systems. We perform an analysis of the demand time series and present evidence of relational dependencies among the stations, motivating the need for a forecasting model that leverages the rich relational structure in the bike sharing networks. Our approach significantly improves multi-step forecasting accuracy of autoregressive time-series models on all three datasets. Further, our approach is easily extendable and we expect applicable to a variety of other time-series forecasting problems.
https://aisel.aisnet.org/hicss-56/da/decision_support_for_scm/3