Yellow Taxi Demand Prediction for New York City Based on
VMD-SSA-LSTM
Haodong Wang
College of Computer Science, Chongqing University, Chongqing 400044, China
Keywords: VMD-SSA-LSTM, Taxi Demand Prediction, New York Yellow Taxi.
Abstract: Taxis have always been a crucial component of urban public transportation systems. Efficient dispatching and
improved operational efficiency are essential for enhancing taxi services. Therefore, accurate prediction of
taxi demand in urban areas is imperative. This paper utilizes a Coupled Network model based on Variational
Mode Decomposition, Sparrow Search Algorithm, and Long Short-Term Memory (VMD-SSA-LSTM) to
predict the demand for yellow taxis in New York City from January to February 2023. The integration of
VMD and SSA proves to be a potent solution to the limitations encountered by traditional LSTM models in
time series analysis, specifically addressing issues of inadequate precision and the intricate nature of
parameter determination. Results from the VMD-SSA-LSTM coupled model show higher accuracy compared
to both traditional LSTM and VMD-LSTM approaches. This indicates that optimized coupled models, such
as VMD-SSA-LSTM, are well-suited for short-term traffic flow predictions. Accurate prediction of taxi
demand facilitates improved scheduling, reduced passenger wait times, increased taxi company revenue, and
contributes to the advancement of smart city initiatives.
1 INTRODUCTION
With the advancement of urbanization, the
coordination between urban transportation and public
transit services has become increasingly crucial. With
urban population growth and an accelerated pace of
life, the demand for taxis in cities has significantly
increased. Therefore, predicting taxi demand holds
significant importance. For the public, forecasting
taxi demand and efficiently dispatching services
make commuting more convenient, reduce waiting
times, and enhance the overall travel experience. For
taxi companies, demand prediction enables rational
scheduling, optimizes resources, increases revenue,
and improves competitiveness. In the context of
urban development, predicting taxi demand
contributes to optimizing traffic management,
fostering economic growth, and promoting the
development of intelligent transportation within a
smart city framework (Cao et al 2021).
Various regions within a city often face situations
where one area experiences a taxi shortage, leading to
long waiting times, while another area has an excess
of taxis, resulting in prolonged idle times (Zhao et al
2019). To avoid this, precise taxi demand prediction
models are essential. Models predicting taxi demand
represent a common form of traffic flow forecasting.
Initially, traffic flow forecasting heavily relied on
mathematical and statistical methods, including
ARIMA models and the K-nearest neighbor
algorithm (Zhang et al 2009). As technology
advances, many machine learning models, including
support vector machines and dynamic Bayesian
networks (Yao et al 2006), have been introduced.
Currently, deep learning methods like Recurrent
Neural Networks (RNN) and Long Short-Term
Memory Networks (LSTM) are extensively used for
traffic flow prediction (Xu et al 2017 & Lai et al
2019).
Due to its capability to generate relatively
accurate forecasts, the LSTM model is commonly
utilized for short-term traffic flow prediction.
However, independent LSTM models have certain
drawbacks, such as insufficiently refined processing
of temporal data and the challenging configuration of
model parameters (Zhao et al 2023). Therefore,
optimizing the LSTM model is essential and
meaningful. Currently, numerous optimized models
for LSTM exist, including VMD-IDBO-LSTM and
SDS-SSA-LSTM (Zhao et al 2023 & Li et al 2022).
This paper adopts a coupled model based on
Variational
Mode Decomposition (VMD), Sparrow