The impact of machine learning forecasting on strategic decision-making for Bike Sharing Systems

Enrico Angelelli; Andrea Mor; Carlotta Orsenigo; M. Grazia Speranza; Carlo Vercellis

math.OCMar 16, 2026

The impact of machine learning forecasting on strategic decision-making for Bike Sharing Systems

Authors: Enrico Angelelli, Andrea Mor, Carlotta Orsenigo, M. Grazia Speranza, Carlo Vercellis

Organizations: Department of Economics and Management University of Brescia, Italy · Department of Management, Economics and Industrial Engineering Politecnico di Milano, Italy

Abstract

In this paper, machine learning techniques are used to forecast the difference between bike returns and withdrawals at each station of a bike sharing system. The forecasts are integrated into a simulation framework that is used to support long-term decisions and model the daily dynamics, including the relocation of bikes. We assess the quality of the machine learning-based forecasts in two ways. Firstly, we compare the forecasts with alternative prediction methods. Secondly, we analyze the impact of the forecasts on the quality of the output of the simulation framework. The evaluation is based on real-world data of the bike sharing system currently operating in Brescia, Italy.

Figures & tables

Figure 1: The structure of the modified simulation framework.

Figure 2: An example of the data defining an instance.

Figure 3: The geographical layout and elevation of the stations of the Bicimia service.

Figure 4: The monthly number of trips by year.

Figure 5: The average number of trips for each half-hour interval by day of the week.

Panel of Figure 6 — Figure 6: Half-hour usage distribution for working days, Saturday, and Sunday.

Panel of Figure 7 — Figure 7: The effect of temperature on the usage of the BSS and the average temperatures in the considered period.

Panel of Figure 8 — Figure 8: The effect of weather phenomena on the use of the system.

Figure 9: The monthly occurrence of each phenomenon as ratio of the days for each month.

Panel of Figure 10 — Figure 10: The trend of the net demand of bikes in different stations.

Figure 11: Usage of the data for the different years.

Figure 12: Distribution of the error by half-hour.

Panel of Figure 13 — Figure 13: Distribution of the error by day of the week: (a) working days, (b) Saturdays, (c) Sundays.

Panel of Figure 14 — Figure 14: The trend of the error in the stations reported in Figure 10 .

Figure 15: Boxplot of the feature importance for the 20 most important features across all models.

Figure 16: Distribution of the average daily missed requests by month.

Figure 17: Distribution of the simulator performance by day of the week in terms of the distribution of total missed requests.

		# Vehicles - Afternoon shift
		0	1	2	3	4	5
# Vehicles - Morning shift	0	145.69	144.77	138.36	132.68	129.11	127.52
	1	131.69	124.91	116.85	112.67	110.72
	2	116.16	108.43	102.84	99.80
	3	106.12	99.48	94.58
	4	101.52	95.22

Table 5: Average daily number of missed requests for each combination of morning and afternoon shifts, when CM is used.

		# Vehicles - Afternoon shift
		0	1	2	3	4	5
# Vehicles - Morning shift	0	145.69	136.43	131.04	128.69	127.39	126.75
	1	122.15	112.56	108.93	107.29	106.30
	2	110.39	102.29	99.02	97.34
	3	103.74	96.43	93.57
	4	100.04	92.52

Table 6: Average daily number of missed requests for each combination of morning and afternoon shifts, when LightGBM-local is used.

Intelligent Transportation Systems

Intelligent Transportation Systems apply sensing, communication, AI, and optimization to make transport networks safer and more efficient.

See all 301 papers

Bikelution: Federated Gradient-Boosting for Scalable Shared Micro-Mobility Demand Forecasting
Feb 24, 2026Antonios Tziorvas, Andreas Tritsarolis, Yannis TheodoridisSpatiotemporal Forecasting Federated Learning
Department of Informatics, University of Piraeus, Piraeus, Greece
Toward Equitable Low-Carbon Mobility: Fairness-Aware Demand Prediction for Expanding Bike-Sharing Systems
Aug 26, 2026Man Luo, Yixuan ZhaoGraph Neural Networks Group Fairness
Department of Computer Science, University of Exeter, Exeter EX4 4QF, U.K.
STAGformer: A Spatio-temporal Agent Graph Transformer for Micro Mobility Demand Forecasting
Jul 7, 2026Ye ZihaoAttention Mechanisms Spatiotemporal Forecasting
Department of System Engineering City University of Hong Kong

Feb 24, 2026J/K move · Enter open · S save

Bikelution: Federated Gradient-Boosting for Scalable Shared Micro-Mobility Demand Forecasting

Antonios Tziorvas, Andreas Tritsarolis, Yannis Theodoridis

Department of Informatics, University of Piraeus, Piraeus, Greece

Spatiotemporal Forecasting Federated Learning Privacy-Preserving ML

Open paper AskPDF arXiv

The rapid growth of dockless bike-sharing systems has generated massive spatio-temporal datasets useful for fleet allocation, congestion reduction, and sustainable mobility. Bike demand, however, depends on several external factors, making traditional time-series models insufficient. Centralized Machine Learning (CML) yields high-accuracy forecasts but raises privacy and bandwidth issues when data are distributed across edge devices. To overcome these limitations, we propose Bikelution, an efficient Federated Learning (FL) solution based on gradient-boosted trees that preserves privacy while delivering accurate mid-term demand forecasts up to six hours ahead. Experiments on three real-world BSS datasets show that Bikelution is comparable to its CML-based variant and outperforms the current state-of-the-art. The results highlight the feasibility of privacy-aware demand forecasting and outline the trade-offs between FL and CML approaches.

Abstract

Intelligent Transportation Systems

Explore similar work

Bikelution: Federated Gradient-Boosting for Scalable Shared Micro-Mobility Demand Forecasting

Toward Equitable Low-Carbon Mobility: Fairness-Aware Demand Prediction for Expanding Bike-Sharing Systems

STAGformer: A Spatio-temporal Agent Graph Transformer for Micro Mobility Demand Forecasting