Exploring the Potentials of Open-Source Big Data and Machine Learning in Shared Mobility Fleet Utilization Prediction

Abouelela, Mohamed; Lyu, Cheng; Antoniou, Constantinos · 2023 · Crossref

DOI: 10.1007/s42421-023-00068-9

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the operational challenges of shared micromobility systems, specifically the difficulty in managing fleet size, vehicle distribution, and demand balancing for shared electric scooters. The rapid, unplanned introduction of these services has created urgent needs for stakeholders to integrate them efficiently into urban transportation networks. To mitigate these issues, the authors developed a practical methodology using open-source big data and machine learning (ML) to predict daily fleet utilization, defined as the number of trips per vehicle. The primary goal was to create a transferable framework that allows operators to forecast demand in a new city based on historical data from an existing city, thereby facilitating dynamic fleet deployment and policy-making. The research employed a model transfer approach, training ML models on long-term historical data from Austin, Texas (the source city), and applying them to predict utilization in Louisville, Kentucky (the target city), which had only limited pilot-stage data. The study utilized four distinct ML algorithms with varying complexity: Linear Regression, Support Vector Regression, Gradient Boosting Machine (specifically LightGBM), and Long Short-Term Memory Neural Networks. Feature engineering incorporated temporal time series features (including neighboring, periodical, and trend information), sociodemographic data, meteorological conditions, and built environment attributes. To address distribution inconsistencies between the two cities, the authors applied sample normalization and label differencing strategies. The results indicated that the Gradient Boosting Machine (LightGBM) achieved the best prediction performance among the tested algorithms. The analysis identified temporal time series features, sociodemographics, weather data, and the built environment as the most critical factors influencing daily fleet utilization. The proposed framework demonstrated high accuracy in predicting the test dataset, validating its potential for real-world implementation. By successfully transferring predictive capabilities from one city to another, the study confirmed that ML models can effectively handle the covariate shift between different urban contexts when proper preprocessing techniques are applied. The significance of this work lies in its provision of a scalable, data-driven tool for shared mobility governance. Unlike traditional regression models, which often exhibit poor predictive power for shared mobility demand, the proposed ML framework offers a robust solution for long-term forecasting. This capability enables operators to dynamically adjust fleet sizes based on predicted demand rather than deploying fixed numbers of vehicles, which can reduce operational inefficiencies, minimize vehicle kilometers traveled during redistribution, and improve service equity. The study contributes to the field by demonstrating the viability of using open-source data and transfer learning to optimize micromobility operations across different urban environments.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction