Volume-3-Issue-10-October,2025 International Journal of Modern Science and Research Technology

ISSN NO-2584-2706

Deep Learning-Powered Route Optimization:

A GIS-Based Approach using Modified Faster R-CNN
Manvi Godbole; Mayuri Rangari;
Yogesh Sonvane; Vidhi Mehta
Dept. of Master in Computer Application, GH Raisoni Collage of Engineering
and Management Nagpur, RTMNU, India

Abstract
Nowadays vehicles are being used a lot classifying traffic congestion levels, and
and different types of vehicles come at a recommending efficient travel routes using
time. Many vehicles running at a time GIS analysis. CCTV footage is used as the
leads to collisions and many people are primary input, with image quality
dying in accidents. Pollution and high enhanced through an Adaptive Gaussian
speed vehicles are running in cities and Bilateral Filter (AGBF) for noise
due to this more than 1.25 million people reduction. A Faster R-CNN model is
are dying every year as a result of road employed for object detection and
traffic accidents globally. Even after that localization within high-risk zones,
no such technology has been available while Particle Swarm Optimization (PSO)
which gives recommendations in real time fine-tunes hyper parameters to boost
to prevent accidents. But my system is detection accuracy. To classify traffic
providing a good solution to this problem congestion into four
which detects the collision of traffic categories—low, moderate, high, and
vehicles and provides collision free route. congestion-free—a CNN-BiGRU
The aim of my research paper is to solve architecture is utilized. GIS further
this challenge by developing an optimal processes spatial and traffic data to suggest
recommendation model on top of a the most efficient, congestion-free routes.
geographic information system in which The model’s performance was evaluated
collision free route is provided using using metrics
advanced planning techniques. Urban such as F1 score, accuracy, precision,
transportation systems face significant recall, and specificity, confirming the
challenges due to traffic congestion and effectiveness of the proposed GINSER
accident-prone zones, leading to increased system. The noise-free images using
delays, environmental pollution, and safety AGBF effectively enhances image quality
concerns. Traditional methods often fall by reducing noise leading to improved
short in delivering real-time, congestion- classification accuracy.
minimized route recommendations and PSO is utilized for hyperparameter tuning
typically do not integrate accident zone achieving a high accuracy of 98.24%. The
detection with traffic management. To GINSER model achieved a classification
overcome these limitations, this study accuracy of 99.16%. The GINSER
introduces GINSER—a Geographic improved overall accuracy by 4.90%,
Information System (GIS)-based optimal 6.83%, 5.23%, and 0.83% better than
route recommendation model powered by TSANet, TCEVis, Ising-traffic, and AID,
advanced deep learning techniques. The respectively. Its ability to detect accident
main goals of the GINSER framework prone zones classify congestion levels and
include identifying accident-prone areas, recommend optimal routes ensures safer
and more efficient mobility.
IJMSRT25OCT082 www.ijmsrt.com 974
DOI: https://doi.org/10.5281/zenodo.18334656
Volume-3-Issue-10-October,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

Keywords algorithms are compared to contemporary

Urban transportation, Traffic classification, optimization techniques [8].The necessity
Deep learning, traffic congestion , R-CNN. for real-time data processing, which is
necessary to detect trends that result in
1.Introduction accidents, limits the effectiveness of the
According to the World Health current approaches [9]. Air quality,
Organization's (WHO) data sheet, traffic economic productivity, and urban mobility
accidents claim the lives of about 1.25 are all adversely affected by the
million people annually. Nearly 400,000 widespread problem of traffic congestion
people under the age of 25 lose their lives [10, 11]. In order to manage the
on the world's roads every year, with over unpredictability of traffic flow, traffic
1000 deaths occurring every day on congestion systems frequently rely on
average.[1Areas on road networks that static solutions like preset routes or fixed
have a high number of traffic accidents traffic signal timings [12]. An intelligent,
due to things like bad road conditions, integrated system that integrates real-time
insufficient signs, a high vehicle density, accident detection, congestion
and reckless driving are known as classification, and optimal route
accident-prone zones. [2 ]Every year, 50 suggestion is desperately needed, as these
million nonfatal injuries occur around the gaps demonstrate [13, 14]. The study
world [3]. These disturbing statistics intends to create the GINSER model using
illustrate Jason GU was the assistant editor GIS, CNN-BiGRU for congestion
who coordinated the evaluation of this classification, and optimized Faster R-
manuscript and approved it for CNN for accident-prone zone detection in
publication. 59134 There is a pressing order to determine the best recommended
demand for improved traffic management route. The CNN-BiGRU model was used
technologies to promote safety and in the suggested method to classify traffic
efficiency in urban transportation in real-time into four levels: low,
networks. The introduction of intelligent moderate, high, and congestion-free.The
transportation systems has resulted in a most effective and traffic-free routes are
desire for intelligent transportation produced by GIS route analysis, which
systems capable of detecting and tracking examines spatial data and traffic patterns.
diverse items, such as vehicles, In places that are prone to accidents, PSO
motorbikes, and buses [4].Congestion is utilized to hypertune the detection and
lengthens travel times, increases fuel localization accuracy of objects.
consumption, and contributes to
environmental pollution. Smart traffic 2.Literature Review
control systems make use of technology Many machine learning and deep learning-
such as IoT, AI, and machine learning. It based studies have been published in
improves traffic flow by analyzing patterns recent years to reduce traffic congestion.
and forecasting congestion in real time [5]. This section examines significant
Deep learning algorithms are critical for advancements in congestion reduction,
detecting accident-prone zones by emphasizing data-driven methodologies
analyzing massive volumes of traffic and and creative control schemes. The
accident data. By utilizing CNN [6]. To following paragraphs provide a list of
anticipate high-risk zones, RNN and deep related studies, with a focus on those that
learning models use data from past use data from traffic congestion
accidents, traffic patterns, weather, and minimization.
road systems [7]. Deep learning is critical Prediction by Abadi et al. (2015) Experts
for lowering traffic congestion in addition in the field of traffic flow analysis and
to accident prediction, and optimization planning can benefit from traffic
IJMSRT25OCT082 www.ijmsrt.com 975
DOI: https://doi.org/10.5281/zenodo.18334656
Volume-3-Issue-10-October,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

prediction to better manage traffic graphically depict their combined impact.

congestion by examining the spatial and Multi-level investigation of road
temporal fluctuations in traffic flow [15]. conditions is made possible by this
Both historical flow data and flow interactive visual analytics solution.
uncertainty were taken into account when In 2024, Kumar et al. [19] proposed the
they created an autoregressive model. This traffic conditions to gauge patterns of
effort was limited, nonetheless, by the lack congestion in situations when there are no
of data to perform further testing in lanes. The spatiotemporal distribution of
incident and normal traffic conditions. nearby road users, such as cars and
When forecasting traffic, special attention motorcycles, defines each traffic state. To
was paid to the complex temporal and record the spatial distribution of nearby
spatial correlations included in traffic data. road users, we use traffic graphs. The
[16]. Zhang et al. (2021) acknowledged effectiveness of TSANet in accurately
the shortcomings of GCN in capturing the predicting traffic conditions across various
changing spatial correlations within the spatial locations within an intersection is
road network and developed a novel model demonstrated by experimental findings on
called the Evolving Temporal Graph the EoT dataset.
Convolutional Network (ETGCN). Within In 2023 Mishra et al. [19] introduced a
a road network, this model sought to visual alert of AP features to drivers using
forecast traffic speeds by learning the dashcam-captured real-time images. On
spatial temporal correlations and their top of several CNN backbones, the
fluctuating states. In 2021, Li and proposed module makes use of attention
Lasenby. In order to shed light on the learning via channel, point, and spatial
XGBoost model and examine how road means. The suggested module’s
and environmental elements affect categorization accuracy can reach up to
accident severity, Li et al. (2022) created 92%. This method has limited scalability
SHAP values. In order to assess the for larger traffic networks and requires
reliability of the elements influencing road high computational resources. The
travel and plan for future development, Ji experimental result of the suggested
et al. (2022) used the SHAP model to module can classify data with an accuracy
calculate the SHAP values of these of up to 92%. In 2024, Kannan [20]
components. Beyond the capabilities of introduced to use of GPS and GIS
conventional geographical statistical technology to analyze spatial patterns of
models, Li (2022) machine learning road accidents, identify high-risk zones,
models can take the role of spatial and suggest targeted preventive measures
statistical models, particularly when in Thanjavur, Tamil Nadu, India. The
handling complex geographic and non- analysis of traffic accidents is done by
spatial aspects. the clustered SHAP police station, vehicle type, time, road, etc.
approach for assessing a variety of firm To take the proper precautions to prevent
capacities, such as profitability and accidents in high-risk locations, an effort is
liquidity, among others (Lin and Gao, also designed to identify the zones that are
2022). A framework was created by Zhang most important. This model relies heavily
et al. (2023) to explain the landslide on the availability of accurate spatial data.
sensitivity assessment models using the In 2024 ElSahly and Abdelfatah [21]
SHAP-XGBoost algorithm [18]. In 2024 introduced the ML-based Traffic Safety for
Dong et al. [18] introduced a machine Automatic Incident Detec tion System. A
learning-based visual analytics system to traffic simulation program is used to
study factors influencing traffic produce realistic, varied traffic data that
congestion. With the multifaceted nature of takes into account variables that have a big
these components, it is difficult to influence on AID performance. 0.89 min

IJMSRT25OCT082 www.ijmsrt.com 976

DOI: https://doi.org/10.5281/zenodo.18334656
Volume-3-Issue-10-October,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

for the (MTTD), 1.01% for the false alarm This study introduces an innovative
rate (FAR), and a detection rate (DR) of GINSER model designed for optimal route
95.6% are obtained from the suggested recommendations utilizing Geographic
training. This computation ally intensive Information Systems (GIS). The input
requires robust simulation environments. images undergo enhanced through
In 2024 Kim et al. [22] introduced pilot Adaptive Gaussian Blur Filtering (AGBF)
zones to confirm the safety of autonomous to improve their quality This initial phase
vehicles (AVs) utilizing data for traffic guarantees that the subsequent detection
accidents. and classification models function on high-
quality data, thereby reducing false
positives and enhancing detection
accuracy. The pre-processing of data is
essential in preparing data for spatial
analysis and machine learning algorithms.
According to many re searchers, crash data
The technique uses a CNN + BiGRU is the most objective and trustworthy
model that was trained using DMV dataset safety data commonly used for road safety
data to determine whether traffic accidents analysis, as it is the most reliable indi cator
were caused by AVs or humans. Accuracy, of road safety. The pre-processed images
Recall, accuracy, and F1 score were used are then analysed using Faster R-CNN,
to evaluate the model’s classification which efficiently detects and localizes
accuracy; the results showed that it objects by integrating region proposals.
achieved 95.03%, 97.8%, 98.9%, and The Region Proposal Network (RPN)
99.5%, respectively. Table 1 presents a generates bounding boxes around regions
comparative literature review of existing of interest (ROI), which are further refined
traffic analysis methods highlighting their by a bounding box regressed for accurate
proposed techniques, advantages, and localization. The hyper parameters of
limitations. In the literature review, these Faster R-CNN are fine-tuned using
existing techniques have several Particle Swarm Optimization (PSO) to
drawbacks like high false positives and ensure robust detection performance.
limited training data due to the rarity of Additionally, CNN-BiGRU is employed to
traffic congestion. Additionally, these evaluate the detected regions, merging
existing routes often fail to provide real spatial feature extraction through CNN
time recommendations for minimizing with temporal dependency analysis. This
congestion, leaving travelers, and planners model categorizes traffic congestion levels
with suboptimal solutions. To overcome into four distinct classes: low, moderate,
these challenges, a novel GINSER was high, and congestion-free. The outputs
introduced for optimal route from both Faster R-CNN and CNN-
recommendation using a geographic BiGRU are subsequently integrated into
information system. the GIS module, which synthesizes spatial
data and traffic patterns to recommend
3.Proposed Methodology optimal routes, particularly in areas prone
Several datasets were used in this study; to accidents and varying congestion levels.
they were cleaned and spatial and data
analysis techniques were applied to
analyze crash rates, predict likelihood of 3.1 Dataset Used Description
road accidents, optimize streets, and find Its Open-source geospatial dataset that
the safest and fastest route between a provides
source and destination in a road network. global geographic data, such as buildings,
roads, landmarks, and natural elements.

IJMSRT25OCT082 www.ijmsrt.com 977

DOI: https://doi.org/10.5281/zenodo.18334656
Volume-3-Issue-10-October,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

This is used to analyse road networks in

Indian cities for identifying traffic
congestion hotspots and accident-zones.
By spatial data on road infrastructure,
traffic density, and historical accident
records, the study identifies critical areas
where congestion and accidents frequently
occur. data offers comprehensive details on
traffic f low, intersections, and route
layouts and land use, which are analyzed
using GIS-based techniques like spatial Figure 2. Pre-processed Traffic Image
overlay, heat maps, and network analysis.
This approach helps urban plan ners and Input – spatial data (S), Traffic
traffic authorities prioritize interventions, Congestion level (T), Accident-prone
such as road redesign, signal optimization, zones(A), Real-time traffic images (I) from
and infrastructure improvements to lessen the user
traffic congestion and improve road. Output-Recommended optimal routes(R) .
Pre-process the input images using AGBF
for noise removal and enhance image
quality.

To detect accident- prone zones using

Faster R-CNN Extract region proposals
with patterns recognize from the images
the traffic data in the main urban area of perform bounding box accurate object
Guiyang city are used to study the urban localization using equation.
traffic congestion zoning boundary To perform optimize Faster R-CNN
problem. In recent years, the urbanization hyperparameter tuning using particle
process of Guiyang city has been swarm optimization
promoted and, according to the data, the
urbanization rate has reached more than
80%; with the number of motor vehicles
reaching more than 5 million, the traffic
congestion in Guiyang city has been
increasing.

Accidents

IJMSRT25OCT082 www.ijmsrt.com 978

DOI: https://doi.org/10.5281/zenodo.18334656
Volume-3-Issue-10-October,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

3.3 Object Detection Using Faster

R-CNN
Now we present a detailed description of
proposed method.
The method is highlighted considering
four steps5 viz (a) minimization with
adaptive background model (b) Faster R-
CNN based subnet operation (c) Faster R-
CNN initial refinement and (d) result
optimization with extended topological
Evaluate fitness to minimize classification active nets
and regression losses
Classify the congestion level using CNN-
BiGRU To integrate GIS for route analysis
and recommendation Use GIS to analyse Faster R-CNN is used
spatial data and recommend routes based
on congestion level accident-prone zones. to detect and localize objects in accident
Recommended the Optimal routes prone zone areas. It builds upon its p
redecessors, R-CNN and Faster R-CNN,
by introducing the region proposal
network (RPN), which significantly speeds
up generating potential object regions.
The ConvNet extracts a rich feature map
from the preprocessed image, which is
then processed by the RPN to propose
regions likely to contain objects. These
regions are refined and

3.2 Pre-Processing
A bilateral filter is a highly efficient
method for denoising digital images while
preserving the edges and detail
information based on geometric distances
and intensity differences between adjacent
pixels. Adaptive Gaussian Bilateral Filter classified into
(AGBF) is used for pre-processing to specific object
improve the image quality and remove categories through a softmax layer after a
noise. completely linked layer classification and
a bounding box regressor for precise
localization.

where (p, q) is the original pixel’s

coordinate in the image to be processed, ‖p
− q‖ spatial distance between pixel p and q
and s represent standard deviation
controlling spatial influence.

IJMSRT25OCT082 www.ijmsrt.com 979

DOI: https://doi.org/10.5281/zenodo.18334656
Volume-3-Issue-10-October,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

Regression Layer

Conv Net
Convoluted feature map

Fully Connected
Layer

3.4 Classification Using Cnn – Bigru dependencies making it particularly

CNN is well-known for using suitable for tasks like text classification.
convolutional layers to capture spatial
hierarchy in data making them highly
effective for feature extraction. The CNN-
BiGRU model combines the strengths of
CNN and bidirectional gated recurrent
units (BiGRU) to process and analyze
sequential data. BiGRU is a recurrent
neural network (RNN) type that is
intended to simulate sequential
dependencies by processing data in both
forward and backward directions. This
combination leverages CNN ability to
extract local patterns and BiGRU GINSER model struggle with the longest
capability to capture temporal route changes due to evolving traffic
conditions highlights the impor tance of
incorporating real-time updates
3.5 Geographic Information System
GIS is a solid framework for collecting,
managing, analyzing, and visualizing
spatial and non-spatial data to understand
geographic patterns and relationships. .
GIS integrates geographic data, mapping,
and spatial analysis algorithms to
recommend the most efficient routes

IJMSRT25OCT082 www.ijmsrt.com 980

DOI: https://doi.org/10.5281/zenodo.18334656
Volume-3-Issue-10-October,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

according on a number of factors,

including time, distance, traffic, and
environmental constraints. It provides an
effective decision-making framework for
urban planning, logistics, transportation
systems, and emergency services by
combining spatial databases, digital maps,
and analytical models. Route
recommendation in GIS involves solving
shortest-path problems, where the goal is
to identify the best route between a source
and a destination. A network is modeled as
a graph G(V, E) , where V is the set of
vertices and E represents edges. Each edge
eij E has a weight wij , representing cost
metrics like distance travel time con
sumption. The most commonly used
mathematical model for route 4.2 Practical Implication
recommendation is the shortest path The user selects CCTV images of traffic
problem where the goal is to minimize the conditions then processed by the GINSER
total cost. model to analyze congestion levels. The
proposed GINSER determines the best
route and provides a congestion-free route
recommendation. This technique leverages
GIS-integrated features that complement
4 Results
traffic management operations making the
This Section utilizing sevel metrics like
model more accessible and applicable to
recall, specificity, accuracy, precision and
real-world scenarios.
F1 score, depending on the images sourced
from the OpenStreetMap datasets. The
4.3 Discussion
results are critical zones, enabling targeted
The GINSER model’s classification
interventions to enhance road safety and
accuracy of 99.16% indi cates remarkable
reduce congestion. The user selected three
dependability in detecting accident-prone
key areas using CCTV cameras: Bhopal,
zones and assessing traffic congestion
Indore, Ujjain. These areas are analyzed to
levels. The metrics like accuracy, recall,
identify accident-prone zones and traffic
and specificity ensure the system limit
congestion hotspots.
false positives and negatives, making it a
dependable decision-making resource. For
4.1 .Performance Analysis
example, the high F1 score throughout
The proposed GINSER method was
congestion levels (99.84% for moderate
evaluated in this section utilizing several
congestion) demonstrates the model ability
metrics, such as F1 score, recall, speci f
to provide fast and accurate
icity, accuracy, and precision, on the
recommendations during peak traffic hours
gathered dataset. Using the mean average
or emergencies. Including PSO increases
precision (mAP), the effectiveness of the
the model's accuracy to 95.24%,
proposed Faster R-CNN detection model
demonstrating its efficacy in decreasing
was assessed.
computational losses and improving the
performance of deep learning approaches
such as Faster R-CNN and CNN-BiGRU.

IJMSRT25OCT082 www.ijmsrt.com 981

DOI: https://doi.org/10.5281/zenodo.18334656
Volume-3-Issue-10-October,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

The GINSER model has essential real- Accid. Anal. Prev. 35 (6), 921–925.
world implications since its GIS-integrated https://doi.org/10.1016/S0001-
frameworks enable dynamic and precise 4575(02)00100-8. Ministry of
traffic manage ment. . Expanding the Energy and Infrastructure (MOEI). 2023,
training datasets to include diverse February 16. Ministry Of Energy And
geographic and demographic contexts will Infrastructure Launches Smart Mobility
improve scalability and ensure broader Think
applicability. Tank To Shape Future Mobility Trends
And Inform Policies. MOEI Official
5. Conclusion Website.
In this research work smart traffic [2] Alkaabi, K.: Identification of hotspot
management of vehicles is studied using areas for traffic accidents and analyzing
Faster R-CNN based deep learning drivers’ behaviors and road accidents.
method. It is an intractable problem in Transp. Res. Interdiscip. Perspect. 22,
computer vision and artificial intelligence 100929 (2023).
domain. Input images are preproc essed [3] V. Adewopo, N. Elsayed, Z. Elsayed,
using AGBF to remove noise and enhance M. Ozer, V. Wangia-Anderson, and A.
images. In accident prone zone areas, Abdelgawad, ‘‘AI on the road: A
PSO-based Faster RCNN is used to detect comprehensive analysis of traffic accidents
and localize objects. In the proposed and accident detection system in smart
model, CNN-BiGRU was used to classify cities,’’ 2023, arXiv:2307.12128..
congestion levels as low congestion, [4] V. Adewopo, N. Elsayed, Z. ElSayed,
moderate congestion, The GINSER M. Ozer, A. Abdelgawad, and M.
enhanced the total accuracy by 3.90%, Bayoumi, ‘‘Review on action recognition
6.71%, 4.13%, and 0.70% better than for accident detection in smart city
TSANet, TCEVis, Ising-traffic, and AID transportation systems,’’ 2022,
respectively. In the GINSER model, arXiv:2208.09588..
external data such as IoT sensors or [5] Mestri, R.A., Rathod, R.R., Garg, R.D.:
weather updates are not integrated, which Identification and removal of accident-
limits its ability to adapt the route prone locations using spatial data mining.
conditions. Finally, we conducted case In: Applications of geomatics in civil
studies using real cab trajectory data to engineering: select proceedings of ICGCE
demonstrate that the approach presented in 2018, Springer, Singapore, 383–394
this paper offered a more comprehensible (2020).
explanation of the potential causes of [6] Geetha, A., Sobia, M.C., Santhi, D.,
traffic congestion. Considering these Ahilan, A.: DEEP GD: deep learning based
limitations, our future work would focus snapshot ensemble CNN with Efficient
on refining the accident detection systems. Net for glaucoma detection. Biomed.
This includes implementing advanced Signal Process. Control 100, 106989
algorithms that can better discern between (2025)
stationary obstacles and genuine accidents, [7] Selvarajan, S.: A comprehensive study
59146 as well as traffic patterns. These on modern optimization techniques for
improvements aim to bolster the reliability engineering applications. Artif. Intell.
and precision of traffic accident detection Rev.. Intell. Rev. 57(8), 194 (2024)
systems in a myriad of real-world [8] Asha, M.B., Stebi, A.A.: Mass Robot:
scenarios. Predictive Maintenance Using Stacked
CNN BI-LSTM for cleaning robots. Int. J.
6. References Data Sci. Artif. Intell. 02(05), 155–160
[1] McKnight, A.J., McKnight, A.S., 2003. (2024) .
Young novice drivers: careless or clueless?

IJMSRT25OCT082 www.ijmsrt.com 982

DOI: https://doi.org/10.5281/zenodo.18334656
Volume-3-Issue-10-October,2025 International Journal of Modern Science and Research Technology
ISSN NO-2584-2706

[9] Dhipa, M., Singh, C.J.C., Michael [16] Li, D., Lasenby, J., 2021.
Priya, E., Shunmugavadivoo, T.: Deep Spatiotemporal attention-based graph
learning-based classification of lichens in convolution network for segment-level
Western Ghats using aerial images. Int. J. traffic prediction. IEEE Trans. Intell.
Comput. Eng. Optim. 02(01), 14–19 Transp. Syst. 23 (7), 8337–8345.
(2024) [17] Li, Z., 2022. Extracting spatial effects
[10] Arun, R., Muthu, B.K., Ahilan, A.: from machine learning model using local
Deep vein thrombosis detection via interpretation method: An example of
combination of neural networks. Biomed. SHAP and xgboost. Comput. Environ.
Signal Process. Control 100, 106972 Urban Syst. 96, 101845.
(2025) [18] Zhang, J., Ma, X., Zhang, J., Sun, D.,
[11] Agoylo, J.C.: GIS-based traffic Zhou, X., Mi, C., Wen, H., 2023. In sights
accident hotspot prediction using machine into geospatial heterogeneity of landslide
learning. Authorea (2024). https:// doi. org/ susceptibility based on the SHAP-xgboost
10. 22541/au.173347433.37543456/v1. model. J. Environ. Manag. 332, 117357.
[12] Chaudhuri, A.: Smart traffic [19] Mishra, S., Rajendran, P.K.,
management of vehicles using faster R- Vecchietti, L.F., Har, D.: Sensing accident-
CNN based deep learning method. Sci. prone features in urban scenes for
Rep. 14(1), 10357 (2024). proactive driving and accident prevention.
[13] Adewopo, V.A., Elsayed, N.: Smart IEEE Trans. Intell. Transp. Syst.Intell.
city transportation: deep learning ensemble Transp. Syst. 24(9), 9401–9414 (2023)
approach for traffic accident detection. [20] Kannan, M.M.: Investigation of road
IEEE Access. (2024). accident zone and suggestion for
https://doi.org/10.1109/ACCESS.2024.338 preventive measures by using GPS and Gis
7972 Technology in Thanjavur, Tamil Nadu,
[14] Zhao, X., Hu, L., Wang, X., Wu, J.: India. Int. J. Eng. Res. Technol. (2024). https://
Study on identification and prevention of doi. org/ 10. 17577/ IJERT CONV1 2IS02034
traffic congestion zones considering [21] ElSahly, O., Abdelfatah, A.: Developing a
machine-learning-based automatic incident
resilience-vulnerability of urban
detection system for traffic safety: promises
transportation systems. Sustainability
and limitations. Infrastructures 9(10), 170
14(24), 16907 (2022).
(2024)
[15] Abadi, A., Rajabioun, T., Ioannou,
[22] Kim, S., Cho, M., Lee, Y.:
P.A., 2015. Traffic flow prediction for road
Exploration of traffic accident-based pilot
transportation networks with limited traffic
zones for autonomous vehicle safety
data. IEEE Trans. Intell. Transp. Syst. 16
validation. Electronis 13(17), 3390 (2024)
(2), 653–662.

IJMSRT25OCT082 www.ijmsrt.com 983

DOI: https://doi.org/10.5281/zenodo.18334656