Functional data analysis approach for mapping change in time series: A case study using bicycle ridership patterns. (January 2023)

• Record Type: Journal Article
• Authors: Roy, Avipsa; Nelson, Trisalyn; Turaga, Pavan
• Is Part Of: Transportation research interdisciplinary perspectives. Volume 17(2023)
• Publication Date: 2023-01
• Subjects: Change detection -- Temporal alignment -- Functional data analysis -- Strava -- Bicycling; DTW Dynamic Time Warping -- FDA Functional Data Analysis -- GPS Global Positioning Systems -- SRVF Square Root Velocity Function; Transportation -- Periodicals; 388.05
• DOI: 10.1016/j.trip.2022.100752
• Languages: English
• ISSNs: 2590-1982

Abstract

Highlights: The manuscript demonstrates a novel framework for change detection in bicycle trips collected from Strava fitness app. Temporal alignment approach is proposed to remove noise from raw timeseries data. Hourly and monthly change maps are created to visualize street-level changes in ridership change patterns. Functional data analysis techniques are implement for time series clustering and temporal alignment. Reproducible methods are created for ease of replicability by policymakers. Abstract: Monitoring change is an important aspect of understanding variations in spatial–temporal processes. Recently, 'big data' on mobility, which are detailed across space and time, have become increasingly available from crowdsourced platforms. New methods are needed to best utilize the high spatial and temporal resolution of such data for monitoring purposes. These data can be considered mappable time series but are challenging to use owing to varying sampling rates and issues of temporal misalignment. We present a methodological framework for change detection from big data captured by crowdsourced fitness app Strava, which addresses misalignment issues in the underlying ridership patterns and maps temporal clusters of bicycling ridership change in the city of Phoenix, AZ between 2017 and 2018 at the street-segment level. Hourly and monthly changes were classified into four clusters for each time period - mapped along with crash density to highlight variations in bicycling ridership. Using spatially and temporally continuous data our study advances the existing approaches to mobility analysis, by using a functional data analysis approach. Our method is reproducible and can be used to expand studies in other cities for monitoring changes directly from crowdsourced ridership data thereby facilitating the decision-making process by practitioners to assess and plan safe bicycle infrastructure.