Forecasting Demand of Shared Bikes Based on ARIMA Model
Yiming Wang
a
International Business School Suzhou at XJTLU, Xi’an Jiaotong-Liverpool University, Suzhou, 215000, China
Keywords: Shared Bike, Demand Prediction, ARIMA.
Abstract: This research uses the ARIMA model to analyse and predict the hourly demand of shared bikes in Seoul, for
the model is a better time series model and is suitable for studying the relationship between hours and bicycle
demand. Firstly, the researcher processes the data and selects the data sets needed for the study from the
original data. Secondly, the researcher conducts ADF testing on the data to detect whether differentiation is
needed. Furthermore, through the ACF and PACF images, the p value and q value are determined. Finally,
this paper plots against the fitted model and predicts five periods backward. The prediction results are
consistent with the trend of the curve based on ARIMA model. The experiment yields hourly changes, which
can help enterprises adjust bicycles’ number deployed in a timely manner. However, this research only studies
the relationship between time and demand for shared bicycles, and do not consider the short-term impact of
other factors, like weather and special events on demand. Subsequent researchers can conduct further research
based on this paper.
1 INTRODUCTION
Currently, many factors affect the demand of shared
bikes. Based on this problem, different scholars have
established different models to predict it. Li et al. (Li
et al., 2024) proposed a Multi-scale Spatiotemporal
Graph Convolutional Network (MSTGCN) model.
They noticed that travel demand has different spatial
dependencies at different scales, which can
effectively mine the multi-scale spatiotemporal
characteristics of shared bicycle demand and offer
direction for the future. Accurately forecast public
transportation travel demand. To accurately grasp the
number around the subway station, Yang and Jin
(Yang and Jin, 2023) proposed a prediction method
based on ridge regression. The demand for bicycles at
the three test sites exhibited a pattern of being higher
on vacations than on working days, according to the
forecast findings. Using data mining approaches,
Sathishkumar et al. (Sathishkumar et al., 2020)
predicted and discovered that the temperature and
hour were thought to be the most significant variables
in the hourly rental bicycle count.
Yang et al. (Yang et al., 2020) improved short-
term demand forecasting for shared bicycle systems
using traffic graph structure data. This method may
a
https://orcid.org/0009-0003-8657-8283
be readily applied to various applications such as
predicting the dynamics of public transportation
systems, and it can be expanded to many current
models that use geographical data. Some restrictions
and potential areas for more research are also
covered. Wei et al. (Wei et al. 2023) considered how
the built environment interacts with demand for
shared bicycles. They employed the Gradient
Boosting Decision Tree (GBDT) model and the
Shapley Additive Explanation (SHAP) method to
forecast demand for shared bicycle travel, analyse the
factors that influence it, and make recommendations
for future developments in the shared bicycle space.
To strengthen the prediction analysis of bicycle
demand, Ramkumar and Saideep (Ramkumar and
Saideep, 2023) proposed a quantum computing
algorithm and used quantum Bayesian network to
anticipate. Compared with the classic algorithm, it
provides calculation acceleration and can speed up
the calculation of the request of shared bikes.
Based on the clustering and prediction analysis of
Dynamic Time Warping (DTW), Le and Leung (Le
and Leung, 2023) performed a spatiotemporal
analysis of shared bicycle demand. Following this,
they employed Bike-share Service (BSS) to carry out
an extensive investigation of the spatiotemporal