International Research Journal of Engineering and Technology (IRJET) e-ISSN: 2395-0056
Volume: 10 Issue: 08 | Aug 2023 www.irjet.net p-ISSN: 2395-0072
© 2023, IRJET | Impact Factor value: 8.226 | ISO 9001:2008 Certified Journal | Page 219
Spatio-Temporal Data Analysis using Deep Learning
Arijeet Singh
1
, Sajid Javid
2
1
Student, Department Of Computer Engineering, Zakir Husain College of Engineering and Technology,
Aligarh Muslim University
2
Student, Department Of Computer Engineering, Zakir Husain College of Engineering and Technology,
Aligarh Muslim University
---------------------------------------------------------------------***---------------------------------------------------------------------
Abstract - This survey aims to present a thorough overview
of the various spatio-temporal data analysis applications of
deep learning techniques. The study examines the widely used
applications of spatiotemporal data analysis, including
transportation, social media events, environmental concerns,
human mobility, action recognition, and other related areas.
Convolutional Neural Networks, Recurrent Neural Networks,
Graph Convolutional Networks, and other neural network
architectures are just a few examples of the deep learning
algorithms and neural network architectures we discuss in this
article for the various spatio-temporal data domains listed
above.
Key Words: Deep Learning, Spatio-Temporal, Neural
Network, Metrics, Network Architecture
1.INTRODUCTION
Spatio-temporal data analytics is an interdisciplinary field
that deals with the analysis and modelling of data that has
both spatial and temporal characteristics. This type of data is
commonly generated in various domains such as
environmental sciences, geography, transportation, and
urban planning, among others. Deep learning, a subset of
machine learning, has received a lot of attention lately
because of its ability to handle large quantities of complex
and multidimensional data. In particular, deep learning has
shown to be highly effective in handling spatial-temporal
data. The following are some benefits of using Deep Learning
models for Spatio Temporal data analysis over more
conventional techniques[1]:
Learning hierarchical feature representations automatically:
Deep Learning models can learn these representations
automatically from the underlying spatiotemporal data.
Effective function approximation capability: If Deep Learning
models have enough layers and neurons, they can fit any
curve and approximate any complex non-linear function.
Perform better with big data: Conventional machine learning
techniques, such as Support Vector Machines and decision
trees, typically outperform big data alternatives on smaller
datasets before reaching a plateau. When more data is
added, deep learning models' performance may continue to
improve.
Types of Spatio-Temporal Data were identified by Wang et
al. in 2022:
1. Event Data: Discrete events that occur in specific
locations at specific times make up event data. An
event's type, place of occurrence, and time of
occurrence can all be used to define it in general
terms.
2. Trajectory Data: Typically, location sensors installed
on moving objects provide trajectory data. The
sensors transmit and record the object's locations
over time at regular intervals.
3. Point Reference Data: A collection of moving
reference points scattered over a specific area and
time period is used to create these measurements.
4. Video Data: A video (sequence of images) can be
categorized as spatio-temporal data type.
Neighboring pixels typically share similar RGB
values in the spatial domain, exhibiting strong
spatial correlations. Succeeding frames exhibit
significant temporal dependence in the temporal
domain.
The highly complex, substantial, and quickly expanding
Spatio-Temporal data, however, continue to present
problems. Some of the challenges include the development of
interpretable models, and fusion of multi-modal spatio-
temporal datasets[1]. We categorize surveyed research
papers with respect to their domain, such as Transportation,
Environment etc. For each category, we put forth the
problem solved by the referenced literature, it’s
methodology, and results in a concise manner. We have
followed a thematic and chronological method for presenting
our survey.
2. LITERATURE REVIEW
2.1 Transportation
Zhou et al., 2022 focuses on the challenge of modelling the
intrinsic correlation of the ST features that were extracted by
the convolutional network apart from the ST aggregation in
traffic data, as predictions made as a result of this may have
biases that affect subsequent transportation planning