Table 2 A comprehensive comparative study of the previous works

[121]

The authors built a ML algorithm to anticipate next year's traffic depending on data from the previous year utilizing libraries like Numpy, Pandas, Matplotlib, OS, Sklearn, Keras, and TensorFlow

Two Kaggle datasets. The date, time, number of vehicles, and number of intersections are all included in one set of data from 2016. The other is the 2017 traffic data which is identical in every respect to the 2016 data

Regression Model

More factors that affect traffic management should be considered

[122]

The authors aimed to simulate traffic signals using a new environment they proposed and referred to as the Simulation of Urban Mobility (SUMO)

N/A

Q-learning RL method

In SUMO, the vehicles that are currently in motion can have their delay times monitored, controlled, and changed

[123]

The main goal was to propose a mechanism for adaptive traffic control

A Road Traffic Prediction Dataset from the Huawei Munich Research Center

Linear Regression, MLP Regressor, Gradient Boosting Regressor, RF Regressor, and Stochastic Gradient Descendent Regressor

The stochastic gradient had an R² value of 0.9, an Explained Variance (EV) value of 0.9, an MAE value of 12.8, a MAPE value of 29%, and an RMSE value of 18

[124]

The authors focused on a key feature of ITSs, the system's capacity to predict lane changes in traffic flows

The U.S. Federal Highway Administration (FHWA) used a program named "Next-Generation Simulation (NGSIM)" to gather a high-fidelity vehicle flow dataset in 2005

SVM, NB, RF, and DT

SVM has the highest accuracy when predicting when a car would switch lanes

[125]

The authors presented a novel fuzzy logic-based framework with a time series analysis for urban traffic volumes forecasting

Data on transportation networks

Interval type-2 fuzzy logic, BP, and SVM

Predictions made with the BP technique and SVM trained using the type-2 fuzzy logic system are more accurate

[126]

The authors aimed at predicting the traffic congestion along a certain road based on past data

Northern England's M62 was the source of 15-min data collection sessions beginning on October 1, 2019 and ending on October 28, 2019

CP tensor decomposition of traffic data

Compared to the cutting-edge rolling-average prediction algorithms, the proposed forecast method significantly beats the competition

[127]

The authors proposed an ML-based Intelligent Traffic Monitoring System (ML-ITMS) to predict congestion at roadside sensors

N/A

SVM and RF

Accuracy 98.6%

[128]

The authors proposed the GSA-ELM to make more precise short-term traffic predictions

Real traffic data for four of Amsterdam Ring Road’s freeways: A1, A2, A4, and A8

ELM

The GSA-ELM model obtains MAPEs of 11.69%, 10.25%, 11.72%, and 12.05%, whereas the RMSEs fall in the 288.09%-163.24% range

[120]

The authors investigated Supervised ML algorithms as a Big Data based analytics technique for traffic volume prediction

The Number of vehicles moving on the road network in the Republic of Serbia recorded by many automatic traffic counters installed during that period

DT, Lazy IBk, RF, Random Committee, and Random Tree algorithms

In the first case study, the M5P algorithm has the best performance, whereas the Lazy IBk method has the best performance in the second study

[129]

The authors proposed a ML technique for predicting travel times to better organize traffic

N/A

GPR

Using non-dispersed algorithms on the input data produced complex trained model. In addition, the proposed technique has to do with the dependence on the quality of the input data

[130]

Hourly traffic forecasting on a section of road in Tangiers, a city in northern Morocco

a dataset containing annualized traffic numbers from 2013 to 2017 that was compiled by the Moroccan Center for Road Studies and Research

SLFN

The proposed model has been evaluated via a comparison with three industry standard algorithms (MLP, SVR, and ARIMA), all of which have concluded that the proposed model offers higher performance in terms of accuracy and consistency

[131]

The authors investigated the power of various ML techniques to predict traffic conditions

Preliminary data was collected over two weeks of monitoring in Bandung, Indonesia

Neural Networks, NB, DT, SVM, DNN, and DL

The size of the training data was very small. In addition, the training process must be reapplied to reflect the newer data set for every change in the training data

[132]

Evaluate the predictive skills of some ML models

Data from the road network in Thessaloniki, Greece

RF, SVR, MLP, and MLR

The SVR model works best in stable situations with small changes, while the MLP model adapts better to situations with more changes and has the fewest errors that are close to zero

[7]

A preliminary method utilizing graphical representations and dimension reduction methods for assessing road traffic accidents was suggested

Realistic road traffic accident data from Gauteng Province, South Africa (SA)

NB, Logistic regression, and K-NN

The resulting performance measures provided some comprehensive insights into the patterns of road traffic accidents

[133]

To predict future foot traffic, the authors presented a new framework based on stepwise regression inside a drift-free setting

Datasets collected from the Caltrans PeMS

Learn++ for SVR, R2C method,

The R2C architecture's prediction of motion volume suffers from a flaw due to its disregard for the spatial dependency of motion volume