A Systematic Literature Review of Learning-Based Traffic Accident Prediction Models Based on Heterogeneous Sources

Abstract

Statistics affirm that almost half of deaths in traffic accidents were vulnerable road users, such as pedestrians, cyclists, and motorcyclists. Despite the efforts in technological infrastructure and traffic policies, the number of victims remains high and beyond expectation. Recent research establishes that determining the causes of traffic accidents is not an easy task because their occurrence depends on one or many factors. Traffic accidents can be caused by, for instance, mechanical problems, adverse weather conditions, mental and physical fatigue, negligence, potholes in the road, among others. At present, the use of learning-based prediction models as mechanisms to reduce the number of traffic accidents is a reality. In that way, the success of prediction models depends mainly on how data from different sources can be integrated and correlated. This study aims to report models, algorithms, data sources, attributes, data collection services, driving simulators, evaluation metrics, percentages of data for training/validation/testing, and others. We found that the performance of a prediction model depends mainly on the quality of its data and a proper data split configuration. The use of real data predominates over data generated by simulators. This work made it possible to determine that future research must point to developing traffic accident prediction models that use deep learning. It must also focus on exploring and using data sources, such as driver data and light conditions, and solve issues related to this type of solution, such as high dimensionality in data and information imbalance.

1. Introduction

The World Health Organization (WHO), through the Global Status Report on Road Safety (GSRRS) 2018, affirms that the number of deaths by road traffic-related issues reached the number of 1.35 million people in 2016 [1]. Meanwhile, the Pan American Health Organization (PAHO) [2] affirms that traffic accidents were the second cause of death among young adults (15–29 years old) in 2016. However, the most concerning is that 47% of all people who died in traffic accidents are vulnerable road users, such as motorcyclists, cyclists, and pedestrians.

The implementation of technological infrastructure and the adoption of strict traffic policies have significantly reduced the accident rate. However, the number of victims is still high and beyond expectations. This situation partly happens because it is complex to determine the real causes of traffic accidents. In most cases, their occurrence depends on one or many of the following factors: mechanical problems, adverse weather conditions, mental and physical fatigue, negligence, potholes in the road, among others.

At present, the use of prediction models as mechanisms to mitigate mortality in traffic accidents is a reality. The results of these models are helping policymakers, transportation safety designers, and researchers to identify factors and make recommendations to make significant achievements in terms of the accident rate [3,4]. Some studies are being funded by institutions or companies related to transportation, as in [4,5,6,7,8,9]. As soon as the prediction model can correlate information from heterogeneous sources, the model might infer accidents in a better way. However, this solution also brings along some issues to be resolved. For instance, some of them are the high dimensionality in data caused by information imbalance or the poor handling of large-scale datasets. In that way, the strategy to improve the prediction models must be focused on exploring other data sources to correlate them and finding strategies to resolve the issues related to this solution.

Since the models are generally fed with real data, the authors have resorted to government platforms and Internet services to collect data. The information from Internet services can be integrated into the prediction model to establish real-time information channels and improve their accuracy. However, this approach is not always feasible because the values and metrics of the different sources are not entirely comparable. In fact, there is much diversity in experimental design, acquisition protocol, equipment used, and data volume. For these reasons, it is important to highlight the current state of the development of learning-based traffic accident predictions and determine the main research challenges on this topic.

This paper presents a systematic literature review on learning-based traffic accident prediction models based on heterogeneous data sources. To elaborate on this review, we used the general guidelines proposed by Kitchenham’s methodology [10,11]. The research questions and search strategy focused on identifying the most relevant features that influence the accuracy and performance of accident prediction models. With this analysis in place, our purpose is to respond to these concerns: How do human factors influence the occurrence of traffic accidents? How does the number of features used in a model affect its performance? How can information from different data sources be correlated? What are the solutions for the challenges that real-time prediction models face? What type of algorithms are best suited for traffic accident prediction models? Moreover, can the best model be determined using only the evaluation metrics? For this purpose, we study the different platforms, services, and simulators used to collect data related to traffic and driver behavior. Regarding the survey of traffic accident prediction models, our work includes a comparative study of models, selection algorithms, evaluation metrics, and the percentage of data used for training/validation/testing. Furthermore, the performance obtained by each model is registered, scored, and analyzed. Following this survey, we aim to find open challenges and research niches in the early prediction of traffic accidents to reduce the death of drivers and passengers.

This article is organized as follows: Section 2 presents the methodology used to elaborate this literature review, followed by Section 3, which introduces the answers to all research questions. Section 4 discusses the most relevant thoughts about learning-based accident prediction. Finally, the conclusions of this literature review are presented in Section 5.

2. Materials and Methods

The current study was performed using the guide for systematic reviews proposed by Kitchenham and others [10,11]. For this study, we have considered the following phases and activities: Planning the Review (Research Questions), Conducting the Review (Search Strategy, Study Selection, Study Quality Assessment, and Data Extraction), and Reporting the Review (Results).

2.1. Planning the Review

Research Questions

In this stage, we present seven research questions developed based on the goals of our research.

• RQ01. What are the data sources used by learning-based traffic accident prediction models?
• RQ02. Where were the datasets used by the prediction models extracted from?
• RQ03. What shortcomings are present in the prediction models?
• RQ04. What are the most common algorithms used by the prediction models?
• RQ05. What are the evaluation metrics used by the prediction models?
• RQ06. What is the performance obtained from the prediction models?
• RQ07. What percentages of the data are used by the models for training, validation, and testing?

2.2. Conducting the Review

2.2.1. Search Strategy

The bibliographic databases and journal platforms used in this review were: Scopus, ACM Digital Library, IEEExplore, Springer Link, and Google Scholar. According to [12], Scopus and Web of Science provide a better quality of indexing and bibliographic records, at least in the computer science field. IEEExplore was picked out because it focuses exclusively on computer science, engineering, and electronics. ACM covers the area of computing and information technology. IEEExplore is considered one of the largest collections worldwide of technical literature. Finally, Springer Link was picked out because it contains many peer-reviewed journals and provides full-text access.

Based on the research questions presented, we extracted the following keywords: real-time, traffic accident prediction, learning, heterogeneous, data source, learning technique, algorithm, and evaluation metric. We added “predicting” and “forecast” to the keyword list as a synonym for prediction. We also developed a list of search strings combining the extracted keywords with the operators “AND” and “OR.” We established three search strings (SS01, SS02, and SS03). SS01 is longer and more specific because it includes all the keywords and synonyms. SS02 does not include the keyword “real-time” from SS01, and SS03 that is less specific, does not include the keyword “heterogeneous” from SS02. This strategy implies that the results returned by each database or platform have duplicate items.

Table 1. Search Results.

Database Search Engine	ID	Command Search	Search Date	Total
Scopus	SS01	ALL(real-time AND “traffic accident*” AND (predicti* OR forecast*) AND learning AND heterogeneous AND “data source*”)		48
	SS02	ALL(“traffic accident*” AND (predicti* OR forecast*) AND learning AND heterogeneous AND “data source*”)	1 April 2021	61
	SS03	ALL(“traffic accident*” AND (predicti* OR forecast*) AND learning AND “data source*”)		154
				263
ACM	SS01	[All:real-time] AND [All:“traffic accident*”] AND [[All:predicti*] OR [All:forecast*]] AND [All:learning] AND [All:heterogeneous] AND [All:“data source*”]		10
	SS02	[All: “traffic accident*”] AND [[All:predicti*] OR [All:forecast*]] AND [All:learning] AND [All:heterogeneous] AND [All: “data source*”]	1 April 2021	10
	SS03	[All: “traffic accident*”] AND [[All:predicti*] OR [All:forecast*]] AND [All:learning] AND [All:"data source*"]		13
				33
IEEExplore	SS01	“Full Text & Metadata”:real-time AND “Full Text & Metadata”:“traffic accident*” AND (“Full Text & Metadata”:predicti* OR “Full Text & Metadata”:forecast*) AND “Full Text & Metadata”:learning AND “Full Text & Metadata”:heterogeneous AND “Full Text & Metadata”:“data source*”		122
	SS02	“Full Text & Metadata”:“traffic accident*” AND (“Full Text & Metadata”:predicti* OR “Full Text & Metadata”:forecast*) AND “Full Text & Metadata”:learning AND “Full Text & Metadata”:heterogeneous AND “Full Text & Metadata”:“data source*”	1 April 2021	136
	SS03	“Full Text & Metadata”:“traffic accident*” AND (“Full Text & Metadata”:predicti* OR “Full Text & Metadata”:forecast*) AND “Full Text & Metadata”:learning AND “Full Text & Metadata”:“data source*”		360
				618
Springer	SS01	real-time AND “traffic accident*” AND (predicti* OR forecast*) AND learning AND heterogeneous AND “data source*”		115
	SS02	“traffic accident*” AND (predicti* OR forecast*) AND learning AND heterogeneous AND “data source*”	1 April 2021	145
	SS03	“traffic accident*” AND (predicti* OR forecast*) AND learning AND “data source*”		352
				612
Scholar	SS01	real-time AND “traffic accident*” AND (predicti* OR forecast*) AND learning AND heterogeneous AND “data source*”	1 April 2021	397
				397

presents the search strings developed for this study and the search results.

2.2.2. Study Selection

Some inclusion and exclusion criteria have been established to accomplish the study selection process.

• Inclusion criteria

  -

  IC01. Published in science, technology, and transportation journals and proceedings;

  -

  IC02. Peer-reviewed research papers;

  -

  IC03. Articles proposing traffic accident prediction models.

• Exclusion criteria

  -

  EC01. Published in health, psychology, or medical journals and proceedings;

  -

  EC02. Literature reviews, mapping studies, chapters in books, theses, technical reports, research proposals, lectures notes, or handbooks;

  -

  EC03. Published in preprint platforms;

  -

  EC04. Articles without full text;

  -

  EC05. Articles proposing traffic accident detection models.

2.2.3. Study Quality Assessment

In this stage, we defined the assessment questions used in the quality instrument. Additionally, we established two or three possible answers for each question and their scores. Thus, the answer “no” with 0 and “yes” is rated with 0.5 or 1.0 depending on the condition. We present the assessment questions and a short justification for them as follows.

The best way to evaluate a model is through the analysis of its evaluation metrics. Since some metrics are more robust and useful than others, having many of them helps to improve the model and its performance.

• AQ01. Does the study present evaluation metrics
  • If the number of metrics = 1, the value is 0.5;
  • If the number of metrics > 1, the value is 1.0.
  Determining the real causes of traffic accidents is complex because they depend on many factors. Thus, the success of such a prediction model lies in correlating different data sources.
• AQ02. Does the prediction model correlate information from different data sources?
  • If the number of data sources = 1, the value is 0.5;
  • If the number of data sources > 1, the value is 1.0.
  Proposing a prediction model by choosing one algorithm and calculating a metric is somewhat imprecise. This process requires an analysis of the model with several baseline algorithms to identify the best one based on indicators and metric values.
• AQ03. Does the prediction model use different automatic learning algorithms?
  • If 0 < the number of algorithms ≤ 2, the value is 0.5;
  • If the number of algorithms > 2, the value is 1.0.
  In general, the prediction models have to deal with high dimensionality and imbalance in information, poor handling of long-scale datasets, or insufficient capacities to process and analyze information. Our study also needs to know the challenges faced by traffic accident prediction models.
• AQ04. Does the study present challenges that the prediction models must face?
  • If the study presents any challenge, the value is 1.0.
  The correct handling of missing and out-of-range data will prevent the occurrence of a bias that invalidates the study. The following studies include missing data treatment in their proposals [13,14,15,16,17].
• AQ05. Does the study include missing data treatment?
  • If the study includes any data missing treatment, the value es 1.0.
  We established, as a selection criterion, that only if the sum of all five questions is greater than or equal to the value defined as the boundary for the first quartile, then the primary study is accepted; otherwise, it is rejected. This value corresponds to 2.5. The research community has widely accepted this selection criterion [11,18].

  Table A1. Quality instrument.

  Title	AQ01	AQ02	AQ03	AQ04	AQ05	Total
  A Bayesian network based framework for real… [4]	1.0	1.0	0.5	1.0	0.0	3.5
  A Bayesian network model for real-time crash… [19]	1.0	1.0	0.5	1.0	0.0	3.5
  A crash-prediction model for multilane roads [54]	0.0	1.0	1.0	0.0	0.0	2.0
  A deep learning approach to the Citywide… [20]	1.0	0.5	1.0	1.0	0.0	3.5
  A genetic programming model for real-time crash… [7]	0.5	1.0	0.5	1.0	0.0	3.0
  A model of traffic accident prediction based on… [21]	0.5	1.0	0.5	1.0	0.0	3.0
  A New Framework of Vehicle Collision… [22]	0.5	1.0	0.5	1.0	0.0	3.0
  A novel variable selection method based on… [9]	1.0	1.0	0.5	1.0	1.0	4.5
  A real-time autonomous highway accident… [23]	1.0	0.5	1.0	1.0	0.0	3.5
  A real-time explainable traffic collision inference… [24]	1.0	1.0	1.0	1.0	0.0	4.0
  A rear-end collision prediction scheme based on… [55]	0.5	0.5	0.5	0.0	0.0	1.5
  A semantic-based classification and regression… [25]	1.0	1.0	0.5	1.0	0.0	3.5
  A spatiotemporal deep learning approach for… [26]	1.0	1.0	1.0	1.0	0.0	4.0
  Accident risk prediction based on heterogeneous… [27]	1.0	1.0	1.0	1.0	0.0	4.0
  Crash prediction based on random effect… [28]	1.0	1.0	0.5	1.0	0.0	3.5
  Data integration and clustering for real time crash… [5]	0.0	1.0	0.5	1.0	0.0	2.5
  Deep dynamic fusion network for traffic accident… [29]	1.0	1.0	1.0	1.0	0.0	4.0
  Evaluating the Performance of Explainable… [30]	1.0	0.5	1.0	1.0	0.0	3.5
  Hetero-ConvLSTM: A deep learning approach to… [31]	1.0	1.0	1.0	1.0	0.0	4.0
  Highway crash detection and risk estimation… [32]	1.0	1.0	0.5	1.0	1.0	4.5
  Highway traffic accident prediction using VDS… [33]	1.0	1.0	1.0	1.0	0.0	4.0
  Intelligent algorithm in a smart wearable device… [56]	0.0	1.0	0.5	0.0	0.0	1.5
  Learning deep representation from big and… [34]	1.0	1.0	1.0	1.0	0.0	4.0
  Operational forecasting of road traffic accidents… [35]	0.5	1.0	0.5	1.0	0.0	3.0
  Predicting crashes on expressway ramps with… [36]	1.0	1.0	0.5	1.0	0.0	3.5
  Predicting motor vehicle crashes using Support… [52]	0.0	0.5	0.5	1.0	0.0	2.0
  Predicting traffic accidents through… [37]	1.0	1.0	1.0	1.0	1.0	5.0
  Prediction of Crash Severity on Two-Lane, Two… [38]	1.0	1.0	1.0	1.0	1.0	5.0
  Real-time crash prediction for expressway… [8]	1.0	1.0	0.5	1.0	0.0	3.5
  Real-time crash prediction in an urban… [6]	0.5	1.0	1.0	1.0	0.0	3.5
  RiskCast: Social sensing based traffic risk… [39]	0.5	1.0	1.0	0.0	0.0	2.5
  Real-time estimation of accident likelihood for… [57]	0.0	1.0	0.5	0.0	0.0	1.5
  Road traffic accidents prediction modelling: An… [58]	1.0	1.0	0.0	0.0	0.0	2.0
  Road Traffic Injury Prevention Using DBSCAN… [59]	0.0	0.5	0.5	1.0	0.0	2.0
  SDCAE: Stack Denoising Convolutional… [40]	1.0	1.0	1.0	1.0	0.0	4.0
  Stack ResNet for Short-term Accident Risk… [41]	1.0	1.0	1.0	1.0	0.0	4.0
  Support vector machine in crash prediction at… [42]	1.0	1.0	0.5	0.0	0.0	2.5
  TA-STAN: A Deep Spatial-Temporal… [43]	1.0	1.0	1.0	0.0	0.0	3.0
  Traffic accident prediction based on deep… [44]	1.0	1.0	1.0	1.0	0.0	4.0
  Traffic accident prediction model using… [45]	0.5	1.0	1.0	0.0	0.0	2.5
  Traffic accident prediction using 3-D… [60]	0.0	0.5	0.5	0.0	0.0	1.0
  Utilizing Machine Learning Models to Predict… [46]	1.0	1.0	1.0	1.0	1.0	5.0

  presents the quality instrument and its results, and Figure 1 presents the phase of Conducting the Review. As observed, 1923 articles were found after performing the search strategy activity. Then, 778 duplicate articles were removed, giving a total of 1145 articles. Once the inclusion and exclusion criteria were applied, 1123 articles were excluded, giving a total of 22 articles. After performing the snowballing technique, 20 articles were added, giving a total of 42 articles. Finally, eight articles were rejected because they did not fulfill the quality criterion. Thus, the number of selected primary studies reached 34 papers.

  Table 2. Selected primary studies.

  ID	Authors	Title	Year	Type
  PS01	Hossain et al. [4]	A Bayesian network based framework for real-time crash prediction on the basic freeway segments of urban expressways	2012	Journal
  PS02	Wu et al. [19]	A Bayesian network model for real-time crash prediction based on selected variables by random forest	2019	Conference
  PS03	Ren et al. [20]	A Deep Learning Approach to the Citywide Traffic Accident Risk Prediction	2018	Conference
  PS04	Xu et al. [7]	A genetic programming model for real-time crash prediction on freeways	2013	Journal
  PS05	Wenqi et al. [21]	A model of traffic accident prediction based on convolutional neural network	2017	Conference
  PS06	Xiong et al. [22]	A New Framework of Vehicle Collision Prediction by Combining SVM and HMM	2018	Journal
  PS07	Lin et al. [9]	A novel variable selection method based on frequent pattern tree for real-time traffic accident risk prediction	2015	Journal
  PS08	Ozbayoglu et al. [23]	A real-time autonomous highway accident detection model based on big data processing and computational intelligence	2016	Conference
  PS09	Liu et al. [24]	A real-time explainable traffic collision inference framework based on probabilistic graph theory	2021	Journal
  PS10	Effati et al. [25]	A semantic-based classification and regression tree approach for modelling complex spatial rules in motor vehicle crashes domain	2015	Journal
  PS11	Bao et al. [26]	A spatiotemporal deep learning approach for citywide short-term crash risk prediction with multi-source data	2019	Journal
  PS12	Moosavi et al. [27]	Accident risk prediction based on heterogeneous sparse data: New dataset and insights	2019	Conference
  PS13	Yan et al. [28]	Crash prediction based on random effect negative binomial model considering data heterogeneity	2020	Journal
  PS14	Paikari et al. [5]	Data integration and clustering for real time crash prediction	2014	Conference
  PS15	Huang et al. [29]	Deep dynamic fusion network for traffic accident forecasting	2019	Conference
  PS16	Parra et al. [30]	Evaluating the Performance of Explainable Machine Learning Models in Traffic Accidents Prediction in California	2020	Conference
  PS17	Yuan et al. [31]	Hetero-ConvLSTM: A deep learning approach to traffic accident prediction on heterogeneous spatio-temporal data	2018	Conference
  PS18	Huang et al. [32]	Highway crash detection and risk estimation using deep learning	2018	Journal
  PS19	Park et al. [33]	Highway traffic accident prediction using VDS big data analysis	2016	Journal
  PS20	Chen et al. [34]	Learning deep representation from big and heterogeneous data for traffic accident inference	2016	Conference
  PS21	Golovnin et al. [35]	Operational forecasting of road traffic accidents via neural network analysis of big data	2020	Journal
  PS22	Wang et al. [36]	Predicting Crashes on Expressway Ramps with Real-Time Traffic and Weather Data	2015	Journal
  PS23	Yuan et al. [37]	Predicting traffic accidents through heterogeneous urban data: A case study	2017	Conference
  PS24	Effati et al. [38]	Prediction of Crash Severity on Two-Lane, Two-Way Roads Based on Fuzzy Classification and Regression Tree Using Geospatial Analysis	2015	Journal
  PS25	Wang et al. [8]	Real-time crash prediction for expressway weaving segments	2015	Journal
  PS26	Basso et al. [6]	Real-time crash prediction in an urban expressway using disaggregated data	2018	Journal
  PS27	Zhang et al. [39]	RiskCast: Social sensing based traffic risk forecasting via inductive multi-view learning	2019	Conference
  PS28	Chen et al. [40]	SDCAE: Stack Denoising Convolutional Autoencoder Model for Accident Risk Prediction Via Traffic Big Data	2018	Conference
  PS29	Zhou et al. [41]	Stack ResNet for Short-term Accident Risk Prediction Leveraging Cross-domain Data	2019	Conference
  PS30	Dong et al. [42]	Support vector machine in crash prediction at the level of traffic analysis zones: Assessing the spatial proximity effects	2015	Journal
  PS31	Zhu et al. [43]	TA-STAN: A Deep Spatial-Temporal Attention Learning Framework for Regional Traffic Accident Risk Prediction	2019	Conference
  PS32	Yu et al. [44]	Traffic accident prediction based on deep spatio-temporal analysis	2019	Conference
  PS33	Sharma et al. [45]	Traffic accident prediction model using support vector machines with Gaussian kernel	2016	Conference
  PS34	Al Mamlook et al. [46]	Utilizing Machine Learning Models to Predict the Car Crash Injury Severity among Elderly Drivers	2020	Conference

  presents the primary studies that were selected.

2.2.4. Data Extraction

We designed four data collection forms to record the selected primary studies’ information. The data collection forms proposed for this section are shown in

Table 3. Classification of algorithms by category.

ID	Algorithms
	Algorithms/Probability Models	Ranking/Variables Selection	Categories
PS01	Bayesian Belief Net	Random Multinomial Logit	Classification
PS02	Bayesian Network	Random Forest	Classification/Ensemble
PS03	Long Short-Term Memory Neural Network		Neural Networks
PS04	Genetic Programming	Random Forest	Evolutionary Computation/ Ensemble
PS05	Convolutional Neural Network		Neural Networks
PS06	Support Vector Machine		Classification
PS07	Bayesian Network	Frequent Pattern Tree/Random Forest	Classification
PS08	K-Nearest Neighbor/Regression Tree/Feed Forward Neural Network		Classification/Ensemble/ Neural Networks
PS09	Bayesian Network		Classification
PS10	Ontology-based Classification and Regression Tree		Classification/Regression
PS11	Convolutional Long Short-Term Memory Neural Network		Neural Networks
PS12	Deep Neural Network		Neural Networks
PS13	Negative Binomial/Random Negative Binomial		Probability Distributions
PS14	Bayesian Network		Classification
PS15	Multilayer Perceptron		Neural Networks
PS16	Gradient Boosting		Ensemble
PS17	Convolutional Long Short-Term Memory		Neural Networks
PS18	Convolutional Neural Network		Neural Networks
PS19	K-Means/Logistic Regression		Clustering/Classification
PS20	Stack Denoise Autoencoder		Neural Networks
PS21	Rumelhart Multilayer Perceptron		Neural Networks
PS22	Bayesian Logistic Regression	Random Forest	Classification/Ensemble
PS24	Fuzzy Classification and Regression Tree		Classification/Regression
PS25	Bayesian Logistic Regression	Random Forest	Classification/Ensemble
PS26	Support Vector Machine/Logistic Regression	Random Forest	Classification
PS27	Multi-view Learning		Not available
PS28	Stack Denoise Convolutional Autoencoder		Neural Networks
PS29	Convolutional Neural Network		Neural Networks
PS30	Support Vector Machine with radial-basis function		Classification
PS31	Deep Learning		Neural Networks
PS32	Long Short-Term Memory Neural Network and Fully Connected Network		Neural Networks
PS33	Support Vector Machine with Gaussian kernel		Classification
PS34	Light Gradient Boosting Machine		Ensemble

,

Table 4. Performance of models.

ID	Evaluation Metrics								AUC	Percentage of Data			Compared with
	MAE	MRE	RMSE	MSE	PAR %	FPR %	TPR %	F1		Train. %	Valid. %	Test. %
PS01					66.00	20.00							Not available
PS02						16.07	70.46						K-Nearest Neighbor, Support Vector Machine, and Logistic Regression
PS03	0.014		0.034	0.001									LASSO and Ridge Regression, Support Vector Regression, Decision Tree Regression, Random Forest Regression, Multilayer Perceptron, and Autoregressive Moving Average
PS04							75.40						Binary Logistic Regression
PS05					78.50					60.0		40.0	Backpropagation Network
PS06					96.70					75.0		25.0	Not available
PS07						38.16	61.11			80.0		20.0	K-Nearest Neighbor
PS08					99.79		42.86						K-Nearest Neighbor/Regression Tree
PS09							0.887	0.813		var.		var.	Artificial Neural Network, Bayesian Regression, and Naive Bayes
PS10			0.267				0.818	0.803	0.807	70.0		30.0	Not available
PS11	0.023			0.019	81.58	0.34							Convolutional Neural Network, Long Short-Term Memory Neural Network, Artificial Neural Network, and Gradient Boosting Regression Tree
PS12								0.590		83.0		17.0	Logistic Regression and Gradient Boosting
PS13	2.520		0.290										Negative Binomial
PS14					76.0								Not available
PS15								0.681	0.786				Support Vector Regression, Logistic Regression, Deep Neural Network, Long-Short Term Memory, and Recurrent Neural Network
PS16					78.00		73.00	0.740		70.0	30.0		Decision Tree and Random Forest
PS17			0.116	0.013						79.0	9.0	12.0	Least Squares Linear Regression, Decision Tree Regression, Deep Neural Network, Fully Connected Long Short-Term Memory, and Convolutional Long Short-Term Memory
PS18					77.34			0.765		80.0		20.0	Convolutional Neural Network
PS19					76.35		40.83			75.0			Logistic Regression and Support Vector Machine
PS20	0.96	0.39	1.0							80.0		20.0	Decision Tree, Logistic Regression, and Support Vector Machine
PS21									0.90				Not available
PS22					90.49		90.40		0.971	70.0	30.0		Not available
PS23					95.12		0.868	0.898	0.961	var.		var.	Support Vector Machine, Decision Tree, and Random Forest
PS24					79.12		0.68						Classification and Regression Tree and Support Vector Machine
PS25					69.80		67.60			70.0	30.0		Not available
PS26							75.03			80.0	20.0		Support Vector Machine
PS27	1.569												Linear Regression /Ridge Regression/Multilayer Perceptron
PS28	0.092	0.796								80.0	20.0		Logistic Regression, Random Forest, Decision Tree, Linear Regression, and Stack Denoise Autoencoder
PS29			0.40	0.16	88.89					87.0		13.0	Auto-Regressive Integrated Moving Average, and Convolutional Long Short-Term Memory Neural Network
PS30					81.3					80.0		20.0	Support Vector Machine with linear
PS31	0.0082		0.0131	0.0001						67.0	11.0	22.0	Linear Regression, Long Short-Term Memory Neural Network, Denoising Auto-Encoder, XGBoost, and Seq2Seq
PS32			0.444		0.723		0.773	0.736		70.0	10.0	20.0	Logistic Regression, Least Absolute Shrinkage and Selection Operator, Support Vector Machine, and Decision Tree
PS33					94.0					70.0	20.0	10.0	Multilayer Perceptron and Support Vector Machine with poly kernel
PS34					87.54		0.814	0.837		80.0		20.0	Logistic Regression, Decision Tree, Random Forest, and Naive Bayesian

,

Table A2. Features list.

ID	Data Source	Type 1	Type 2	Variables
PS01	traffic flow	real	variable	vehicle speed, number of vehicles…
	traffic accidents	real	variable	date, time, location, vehicles involved…
PS02	traffic flow	real	variable	vehicle speed, flow, and occupancy
	traffic accidents	real	variable	time, location, and collision description
PS03	traffic accidents	real	variable	time and location
PS04	weather conditions	real	variable	clear and adverse
	traffic flow	real	variable	vehicle speed, number of vehicles…
PS05	weather conditions	real	variable	sun, cloud, rain, snow, fog, sleet…
	traffic flow	real	variable	vehicle speed, number of vehicles…
PS06	weather conditions	simulated	variable	rain, snow, and fog
	light conditions	simulated	variable	sun, headlights, and streetlight
PS07	weather conditions	real	variable	type and visibility
	traffic accidents			Not available
	traffic flow	real	variable	vehicle speed, volume, and occupancy
PS08	traffic accidents	real	variable	date, time, location…
	traffic flow	real	variable	date, time, number of vehicles…
PS09	weather conditions	real	variable	tweets (snow, sleet, fog…)
	traffic accidents	real	variable	time, street name, location…
	traffic events	real	variable	tweets (closures, incidents…)
PS10	vehicle data	real	static	type and seat belt
	driver’s data	real	static	age, gender, and education level
	weather conditions	real	variable	visibility
	light conditions			Not available
	traffic accidents	real	variable	time, day of week, severity…
	road infrastructure	real	static	geometric characteristics
	others	real	static	topographic map…
PS11	weather conditions	real	variable	average temperature, precipitation…
	traffic accidents	real	variable	date, time, location, collision type…
	taxi trips	real	variable	pick-up timestamp, pick-up location…
	traffic flow	real	variable	volume
	road infrastructure	real	static	road length, road type, and intersections
	population	real	static	Not available
	others	real	static	land use
PS12	weather conditions	real	variable	temperature, pressure, humidity…
	traffic events	real	variable	collision, broken vehicle, congestion…
	road infrastructure	real	static	warning, priority, information…
PS13	weather conditions	real	variable	visibility
	traffic accidents	real	variable	number of property damage only…
	traffic flow	real	variable	average speed limit…
	road infrastructure	real	static	road length, curvature…
PS14	weather conditions	simulated	variable	street identifier, temperature, snow…
	traffic accidents	simulated	variable	time, location…
	traffic flow	simulated	variable	Not available
	road infrastructure	simulated	static	geometric characteristics
	points of interest	simulated	static	location
PS15	traffic accidents	real	variable	timestamp, location…
	traffic events	real	variable	timestamp, location, and category
	points of interest	real	static	place, category, and location
PS16	weather conditions	real	variable	Not available
	traffic accidents	real	variable	date, time, city, state…
PS17	weather conditions	real	variable	precipitation, temperature…
	traffic accidents	real	variable	time and location
	road infrastructure	real	static	speed limits and volume
	others	real	static	satellite images
PS18	traffic accidents	real	variable	identifier, timestamp, location…
	traffic flow	real	variable	volume, average speed, and occupancy
PS19	weather conditions	real	variable	Not available
	traffic accidents	real	variable	time, day, location, number of dead…
	traffic flow	real	variable	time, number of lanes, volume, density…
	road infrastructure	real	static	road shape and alignment
PS20	traffic accidents	real	variable	time, location, and security level
	pedestrian mobility	real	variable	identifier and location
PS21	weather conditions	undefined	variable	categories
	light conditions	undefined	variable	categories
	traffic accidents	undefined	variable	time, day of week, and collision month
	traffic flow	undefined	variable	type and state of the control device…
	road infrastructure	undefined	static	speed limit, road type, pavement type…
	traffic events	undefined	variable	type
PS22	traffic accidents	real	static	time, location, vehicles involved…
	traffic flow	real	variable	average speed, volume, average…
	road infrastructure	real	static	road type, road length, tolls…
	weather data	real	variable	visibility and road surface
PS23	weather data	real	variable	precipitation, temperature…
	traffic accidents	real	variable	time and location
	traffic flow	real	variable	speed limits and annual average daily…
	population	real	static	Not available
	others	real	static	area size of census blocks
PS24	vehicle data	real	static	type
	driver’s data	real	static	age, gender, education level…
	weather conditions	real	variable	sun, fog, rain, snow, storm, dry, wet…
	light conditions	real	variable	day or night
	traffic accidents	real	variable	time, day of week, and vehicle failure
	road infrastructure	real	static	road width, imperfections…
	others	real	static	topographic map and digital elevation…
PS25	weather conditions	real	variable	type, wind direction and speed…
	traffic accidents	real	variable	time, location, collision type…
	traffic flow	real	variable	number of vehicles, occupancy…
	road infrastructure	real	static	entrance and exit ramp
PS26	vehicle data	real	static /	identifier, time, vehicle speed, and type
			variable
	traffic accidents	real	variable	date, time, location, and collision type
PS27	traffic accidents	real	variable	tweets and police report
PS28	traffic accidents	real	variable	identifier, time, location…
	traffic flow	real	variable	device identifier, timestamp…
PS29	weather conditions	real	variable	precipitation, snowfall, temperature…
	road infrastructure	real	static	number of lanes, road type…
	pedestrian mobility	real	variable	people’s arrivals and departures
	points of interest	real	static	place, location, and category
	population	real	static	Not available
	others	real	static	weekends and holidays
PS30	traffic accidents	real	variable	number of dead and victims…
	road infrastructure	real	static	road width, segment length…
	population	real	static	Not available
	others	real	static	geographical area, income…
PS31	vehicle data	real	variable	location, pick up and pick off time
	weather conditions	real	variable	date, time, location, temperature…
	traffic accidents	real	variable	time, place, street, collision reason
	road infrastructure	real	static	geometric characteristics
	taxi trips	real	static	taxi zones
	points of interest	real	static	name, location, and category
	others	real	variable	start and end point
PS32	weather conditions	real	variable	temperature, dew point, humidity…
	traffic accidents	real	variable	time and location
	road infrastructure	real	static	name and points for roads…
	taxi trips	real	variable	time, location, and speed
	points of interest	real	static	name, location, and category
PS33	vehicle data	simulated	variable	speed and vehicle condition
	driver’s data	simulated	static	age, involvement of alcohol and drugs
	weather conditions	simulated	variable	Not available
PS34	traffic accidents	real	variable	age, gender, injury, collision year…
	traffic flow	real	variable	volume
	road infrastructure	real	static	geometric characteristics (speed limits)

and

Table A3. Datasets and simulators.

ID	Datasets/Services	Simulators
PS01	Metropolitan Expressway Company Limited and Vehicle Collision and Normal Traffic Condition [61]
PS02	California Department of Transportation and Highway Performance Measurement System [62]
PS03	Beijing traffic accident data
PS04	The Statewide Integrated Traffic Records System [63], Highway Performance Measurement System [64], and Interstate 880 Highway
PS05	Interstate 15 Highway
PS06		Prescan [65] and Matlab/Simulink
PS07	Virginia Department of Transportation [66] and Interstate 64 Highway
PS08	Intelligent Transportation Systems and Real-Time Monitoring System [67]
PS09	Twitter API [68] and New York City Open Data [69]
PS10	National Cartographic Centre [70], Ministry of Roads and Urban Development [71], Meteorological Organization [72], and Highway Police [73]
PS11	New York of Police Department (Vehicle collisions) [69], New York City Taxi and Limousine Commission, Taxi GPS Data [74], New York City Department of Transportation [75], United States Census Bureau (TIGER files) [76], New York City Department of City Planning [77], and National Climatic Data Center [78]
PS12	US-Accidents dataset [79], MapQuest Traffic [80], and Microsoft Bing Map Traffic [81]
PS13	Washington State Department of Transportation [82], Highway Safety Information System [83], and Digital Roadway Interactive Visualization and Evaluation Network [84]
PS14		Paramics Microsimulation [85], AIMSUN [86], and VISSIM [87]
PS15	New York Police Department (Traffic Accident Dataset) [69], Points of Interest from New York City [88], and New York City’s governmental platform [88]
PS16	US Accidents (A Countrywide Traffic Accident Dataset) [89] and The Weather Channel [90]
PS17	Iowa Department of Transportation [91], Iowa Department of Transportation (RWIS) [92], Iowa Department of Transportation (Iowa DOT GIS) [93], and Google Earth Satellite Image [94]
PS18	Iowa Department of Transportation (Traffic Management Centers Reports) [91], Iowa DOT (Interstate 235 (I-235) and Traffic Flow) [91]
PS19	Korea Expressway Corporation (Traffic Flow) [95] and Korean National Policy Agency (Traffic Accidents) [96]
PS20	Japan traffic accident data and Japan human mobility data
PS21	Not available
PS22	Signal Four Analytics [97], Central Florida Expressway Authority [98], and National Climatic Data Center (Weather Data) [78]
PS23	Iowa Department of Transportation (Vehicle collisions) [91], Stage IV radar rainfall [99], Iowa Department of Transportation (RWIS) [92], Iowa Department of Transportation (Iowa DOT GIS) [93], and Census Data [76]
PS24	Iran National Cartographic Center [70], Ministry of Roads and Urban Development Islamic Republic of Iran [71], Iran Meteorological Organization [72], National Geographical Organization of Iran [100], and the Information and Technology Department of the Iranian Traffic Police [73]
PS25	Signal Four Analytics [97], National Climatic Data Center [78] and Central Florida Expressway Authority [98]
PS26	Autopista Central [101] and Department of Geophysics of University of Chile [102]
PS27	New York City Police Department (Public Traffic Accident Report) [103]
PS28	Xiamen traffic accident data and Vehicle License Plate Recognition sensors
PS29	New York City data
PS30	Florida Department of Transportation (Crash Analysis Reporting System) [104], FDOT (Roadway Characteristics Inventory) [104], Map of Hillsborough, and United States Census Report [76]
PS31	New York of Police Department (Vehicle collisions) [69], New York City Taxi and Limousine Commission (Trip Data) [74], National Climatic Data Center (Weather Data) [78], and New York City Open Data [69]
PS32	Beijing’s datasets about traffic accidents and Weather Underground [105]
PS33	Questionnaires filled by drivers, pedestrians, and others
PS34	Office of Highway Safety Planning (Michigan Traffic Crash Facts Dataset) [106]

. The design of them was based on addressing the research questions. Thus, Table A2 was designed to answer RQ01, Table A3 to answer RQ02, Table 3 to answer RQ04, and Table 4 to answer RQ05, RQ06, and RQ07. Table A2 includes the primary study ID, the data sources (vehicle data, driver’s data, weather and light conditions, traffic accidents, traffic flow, traffic events, road infrastructure, taxi trips, points of interest, and others), two categories to refer to the data type, and a list of variables of features of each data source. Table A3 includes the primary study ID, the datasets, services, or simulators. Table 3 includes the primary study ID, the algorithm or algorithms used on the model, and the groups to which those belong [47,48]. Finally, Table 4 includes the primary study ID, some evaluation metrics, percentages of data used for training, validation, and testing, and the algorithms used by models to compare their performance. The generated data will be presented in the “Results” section and analyzed and interpreted in the “Discussion”.

3. Results

3.1. Study Overview

Considering the year and the type of publication (Table 2), from 34 selected studies, 19 of them are articles from journals and 15 of them from conferences. The years in which more papers were published were 2015, 2018, and 2019. The answers to our research questions are presented as follows.

3.2. RQ01

The prediction models use the following data sources: vehicle data, driver’s data, weather conditions, light conditions, traffic accidents, traffic flow, traffic events, road infrastructure, taxi trips, points of interest, and population. The most common data sources are weather conditions, traffic accidents, traffic flow, and road infrastructure. Meanwhile, driver’s data, light conditions, and taxi trips are the least common. Based on Table A2, the attributes contained in each data source are presented as follows.

• Vehicle data: identifier, time, location, type, speed, condition, seat belt, pick up and pick off time;
• Driver’s data: age, gender, education level, collision factors (sleepiness and boredom), and involvement of alcohol and drugs;
• Weather conditions: sun, cloud, rain, snow, fog, sleet, crosswind, sand, dawn, dusk, visibility, temperature, precipitation, snowfall, pressure, wind speed, humidity, hail, storm, wind direction, and dew point;
• Light conditions: headlights, streetlights, sunlight, and night light;
• Traffic accidents: vehicles involved, collision type, collision description, the direction of the road, number of killed or injured people, severity, human situation, number of property damage only collisions, number of collisions with casualties and dead, presence of traffic objects, road segment, event type, security level, collision month, vehicle failure, police report, and origin of the collision;
• Traffic flow: vehicle speed according to radar, number of vehicles, occupancy, average speed, annual average daily traffic, driving direction, and lane identifier;
• Traffic events: closures, constructions, broken vehicles, collisions, congestion, and blocked lanes;
• Road infrastructure: geometric characteristics (road length, road shape, road alignment, road type, number of lanes, horizontal curve radius, width of shoulder, slope, tunnel, imperfections, intersections, entrance and exit ramp, and speed limits), and road signs (warning, priority, information, facilities, and service);
• Taxi trips: pick-up and drop-off timestamp, pick-up and drop-off location, trip distance, payment information, taxi zones, and taxi speed;
• Points of interest: place, category, and location;
• Population: *not shown;
• Other: topographic map, digital elevation map, land use, satellite images, the area size of census blocks, special calendar dates, geographical area, trip survey, and bike trip.

3.3. RQ02

The prediction models are fed with data collected from open and government platforms, others from Internet services, and even others with simulators’ data. According to Table A3, the platforms, Internet services, and simulators used by the models to collect data are presented as follows.

• Open platforms: Kaggle and Open Data;
• Government platforms: Institutions of statistics and census, geographical and meteorological organizations, and departments of police and transportation;
• Internet services: MapQuest Traffic, Microsoft Bing Map Traffic, The Weather Channel, Weather Underground, Google Earth Satellite Image, and Twitter;
• Simulators: AIMSUN, VISSIM, PreScan, and Paramics Discovery;
• Applications: Intelligent Transportation Systems (ITS) and Real-Time Monitoring Systems;
• Others: Questionnaires.

3.4. RQ03

Considering that “no model is perfect”, the prediction models present at least some of the following shortcomings.

• Non-inclusion of spatial heterogeneity within the zones of study;
• Information imbalance (the amount of useless data is greater than useful data) because most data are non-accident related;
• Insufficient capacities to process and analyze an enormous amount of data;
• Poor handling of long-scale datasets. It is not practical to work with huge amounts of raw data; therefore, it is necessary to select relevant features to be extracted. If this selection is not made adequately, the generated models will not work correctly;
• Not having enough related information to train and test the models (e.g., it is essential to have information about traffic accidents and normal traffic conditions from the same segment).

3.5. RQ04

The most common algorithms among prediction models in order of occurrence are Neural Networks (Long Short-Term Memory NN, Convolutional NN, Deep NN, and Feed Forward NN), Support Vector Machine, and Bayesian Networks. According to Figure 2, 30% of prediction models use some variants of Neural Networks, 15% of them use Support Vector Machine, and 12% use Bayesian Networks. Regarding ranking and selection variables/features, the most common algorithm is Random Forest. The categories to which those algorithms belong are Neural Networks, Classification, and Ensemble. Finally, the most common algorithms used by models to compare their performance are Logistic Regression, Support Vector Machine, Decision Tree, and some variants of Neural Networks. Their categories are Classification and Neural Networks.

3.6. RQ05

The evaluation metrics used by authors are:

• For classification problems: Prediction Accuracy Rate (PAR)/Accuracy, True Positive Rate (TPR)/Sensitivity/Recall, False Positive Rate (FPR)/Fall-Out, F1 Score, and Area Under Curve (AUC);
• For regression problems: Mean Absolute Error (MAE), Mean Relative Error (MRE), Root Mean-Square Error (RMSE), and Mean Squared Error (MSE).

Of all these metrics, the more commonly used are:

• For classification: PAR, TPR, and F1 Score;
• For regression: RMSE and MAE.

3.7. RQ06

The prediction models obtained the results presented as follows. Figure 3 shows the dispersion of values of evaluation metrics.

• Accuracy (%):
  PS01 ≫ 66.00  PS05 ≫ 78.50  PS06 ≫ 96.70  PS08 ≫ 99.79  PS11 ≫ 81.58
  PS14 ≫ 76.00  PS16 ≫ 78.00  PS18 ≫ 77.34  PS19 ≫ 76.35  PS23 ≫ 95.12
  PS24 ≫ 79.12  PS25 ≫ 69.80  PS29 ≫ 88.89  PS30 ≫ 81.30  PS33 ≫ 94.00
  PS34 ≫ 87.54
• Sensitivity (%):
  PS02 ≫ 70.46  PS04 ≫ 75.40  PS07 ≫ 66.11  PS26 ≫ 75.03
• F1 score:
  PS09 ≫ 0.813  PS10 ≫ 0.803  PS12 ≫ 0.590  PS15 ≫ 0.681
• Root Mean-Square Error:
  PS03 ≫ 0.034  PS13 ≫ 0.290  PS17 ≫ 0.116  PS32 ≫ 0.444
• Mean Absolute Error:
  PS20 ≫ 0.960  PS27 ≫ 1.569  PS28 ≫ 0.092  PS31 ≫ 0.008
• Area Under Curve:
  PS21 ≫ 0.900

According to Figure 3, there are four groups of values for PAR, F1, AUC, and MSE. All PAR values range from 0.65 to 1.0 (65% to 100%). Similarly, all F1 and AUC values range from 0.58 to 0.90 and 0.80 and 0.97, respectively. Additionally, all MSE values are located under 0.17. These ranges could be seen as a reference for new models that use these evaluation metrics. By contrast, the values of the rest of metrics are so dispersed that it is not possible to identify group of values to serve as references.

3.8. RQ07

Most models only use data for training and testing; however, a few models also use data for validation. The percentages established by the models are as follows:

• For training: [60.0–83.0]%
• For validation: [9.0–30.0]%
• For testing: [10.0–40.0]%

Even there are models in which those percentages are variable and defined dynamically. The most common split configuration among proposals is 80% for training and 20% for testing.

4. Discussion

Below, we mention some thoughts presented in the articles to analyze and consider for future research. For instance, traffic accidents are not fortuitous events but events caused by conditions that occur in space and time and under certain circumstances [30]. According to [32], unfavorable traffic characteristics, adverse weather conditions, and driver distraction may lead to a crash. Additionally, the most significant factors on crash severity are vehicle failures, not wearing the seat belt, and unfavorable weather conditions [38]. Meanwhile, others assert that driving drunk and at high speed are serious factors in traffic accidents [45], and the wet pavement is one condition that increases the accident rate significantly [8]. Finally, the situation that causes the highest probability of suffering a traffic accident is the aggressive driving behavior after unusual congestion to recover the time lost [6]. For their part, the authors of [25] determined as follows: high speed is one of the most recurrent causes among fatal vehicle crashes; the traffic during morning peak and the first days of the week increase the risk of property-damage-only crashes; additionally, slopes and proximity to curves are the main road geometry factors that lead to fatal crashes; high speed and proximity to curves are the main causes of fatal-injury type crashes; faulty windshield wipers in rainy weather conditions and not wearing seat belt among young people are the most important causes of injury crashes; and, finally, driving at night without caution during rainy weather conditions increase the risk of property-damage-only crashes.

About performance, models based on Deep Neural Networks reduce their accuracy, precision, and F1 score as the learning data size increases [37]. Additionally, the performance of a Support Vector Machine model depends on the learning process, so future efforts must focus on tuning the scale of parameter values and kernel functions selection [42]. Finally, the authors of [26] assert that the performance of the prediction models decreases as the spatio-temporal resolution of the prediction task increases. Regarding features, incorporating more features into the model does not always improve its performance [44]. Meanwhile, ref. [45] asserts that a lesser number of features affect the performance of a neural network. Finally, and according to [37], removing features from models based on Decision Tree or Random Forest has an enormous impact, but slight in models based on Deep Neural Networks.

Some authors propose some recommendations; for instance, splitting data into pieces to send them to compute nodes can make the computational time much lower, which would benefit the handling of social media data [24]. For their part, ref. [19] suggests that the threshold used to separate different states (crash/non-crash) must compensate the values of True Positive Rate and False Positive Rate, and also that the optimal threshold may be found by comparing the performance of different thresholds. Finally, the authors of [5] proposes that the outcomes of a real-time traffic accident prediction model are shown through a variable message sign or transmitted between vehicles using a connected vehicular system.

Despite the advantages that simulators offer at present, this mechanism of data generation has not been received as expected. In fact, there is a clear trend in prediction models about using real data instead of simulated ones. From the results, we could remark that only 1 out of 10 models use data generated by simulators. Because traffic accidents are events caused by a group of conditions that are not always the same and take place in space and time and under certain circumstances, it can be suspected that the authors prefer less controlled scenarios than those provided by simulators to generate data. Moreover, it was noted that there are both static and variable data. Static data, such as most driver data, road infrastructure, points of interest, or satellite images could be used to build a base model. In contrast, data that vary over time, such as traffic accidents, weather conditions, or traffic flow, could be used to adjust the model.

The human factor is the leading cause of traffic accidents [49,50], and the most common human factor (contributing or principal) is inattention while driving because of overloading attention, distraction, or monotonous driving [51]. According to [46], young people are more susceptible than adults to suffer a traffic accident; male drivers are more involved in traffic incidents than female drivers, and female drivers are more susceptible than male drivers to suffering severe injuries. It is clear that the human factor influences and plays an essential role in the occurrence and severity of traffic accidents. This affirmation is confirmed in the Global Status Report on Road Safety. It establishes that factors associated with road user behavior, such as speeding and drink-driving, are two of the key risk factors to be considered and reinforced within the legislation of countries to prevent deaths and injuries due to traffic accidents. Some countries, especially high-income ones, have reduced the number of deaths and injuries by adopting policies for all the key risk factors [1]. Although we have improved much in the prevention of traffic accidents, it is clear that we must now focus on the field of the prediction of traffic accidents. In the context of our research, we could notice that very few models use driver’s data, although the human factor is one of the leading causes of traffic accidents. We believe that this may be due to the non-availability of this type of information.

Considering that the prediction models are generally fed with real data, the authors have resorted primarily to governmental institutions related to transportation or related areas and secondly to Internet services. The information collected from government platforms is mainly related to traffic accidents, traffic flow, and road infrastructure. The information collected from government platforms is mainly related to traffic accidents, traffic flow, and road infrastructure. Internet services provide information mainly related to weather and light conditions and traffic events. Most Internet services (MapQuest Traffic, Microsoft Bing Map Traffic, or Twitter, among others) provide APIs that can be integrated into the model to establish real-time or deferred information channels.

One of the most challenging issues for the traffic accident prediction models is to count on a real-time solution. According to some authors [4,9,33,37], the development of a real-time decision-making tool to avoid traffic accidents is completely viable as soon as shortcomings such as the non-integration of spatial heterogeneity, the incorrect handling of long-scale datasets, the improper handling of unique data properties, the information imbalance, and the lack of related information, are resolved. The correct handling of long-scale datasets requires feature extraction and imbalance correction. First of all, it is not practical to work with a huge amount of raw data, therefore to handle adequately large datasets, it is necessary to extract essential features such as weather, type of environment (for instance, rural highway vs. urban street), road conditions, speed limit, type of traffic, driver data, and type of vehicles [26]. Additionally, the accident-related data are less frequent than the non-accident-related information. Therefore, the datasets are imbalanced, and a predicting model has to be built to correct this situation [32].

It is also essential to consider which characteristics are time-sensitive, time-insensitive, and related to spatial heterogeneity. Time-insensitive data are fully connected, and spatial heterogeneity is a trainable component. It would be possible to obtain a somewhat generalized solution trained for different scenarios starting with a common base that considers this feature differentiation type [52]. These data-handling strategies could make it possible to obtain a real-time prediction which is the next big challenge for this research area.

The high-dimensionality problem may be solved using data processing techniques to derive relevant features through methods, such as clustering, chi-square, Minimum-Redundancy-Maximum-Relevance (mRMR), and predictor importance, among others. Some authors, such as [28], have worked on this strategy for dimensionality reduction using clustering, but other pre-processing techniques could also be tested.

Regarding algorithms, we were able to identify two stages for which machine learning algorithms were assigned. The pre-processing stage includes the tasks of ranking and selecting features, while the classification stage includes the selection of the model. For pre-processing, the most common algorithm is Random Forest; and, for classification, the most common algorithms are some variants of Neural Networks (Long Short-Term Memory NN, Convolutional NN, Deep NN, and Feed Forward NN). This algorithm selection is consistent with the fact that deep learning models applied in the area of Traffic Accident Prediction are becoming more popular. Most authors use shallow learning algorithms as baseline algorithms to compare the performance of their models based on neural networks. This tendency marks a path for research in learning-based accident prediction.

The metrics more commonly used for classification problems are accuracy, sensitivity, and F1 Score; meanwhile, for regression problems, Mean Absolute Error and Root Mean-Square Error. However, there is such a diversity of experimental design, data volume, and structure used in the various studies that it is difficult to compare results using simply evaluation metrics. Not to mention that some proposals present non-normalized values for their evaluation metrics. The datasets are typically unbalanced, and performance must be understood in a contextualized way. Therefore, to compare models to find the one with the best performance is not necessarily real because the results are not completely comparable among the studies.

Although there is no precise rule to split data for training, validation, and testing, a tacit agreement establishes an approximate data split configuration. From the analysis, we could establish that a higher percentage (more than 50%) of data are used for training and a lower percentage (less than 50%) for testing. The most common data split configuration among proposals is 80% for training and 20% for testing. Some models even establish a low percentage of data for validation. It was noted that there is no evidence or justification for splitting data in one way or the other or whether such a data split configuration could improve the performance of the models. Because of this drawback, we could suggest using data splitting methods (e.g., SPlit [53]) instead of splitting randomly to obtain the optimal configuration.

5. Conclusions

The elaboration of this work has made it possible to present a review of the research done so far on learning-based traffic accident prediction. Some of the most important points to be considered are as follows.

The development of prediction models in real-time is viable as soon as issues, such as the efficient use of large-scale datasets, the integration of spatial heterogeneity, and the solution for high dimensionality in data, are resolved. In this context, some solutions for these issues are presented as follows. The efficient handling of large-scale datasets may be solved using feature extraction and imbalance correction; meanwhile, the high dimensionality in data may be solved using data processing techniques.

There is a trend about using real data generated by less controlled scenarios (as in real life) instead of data generated by simulators. Thus, authors have opted to correlate real data usually collected from open and government platforms with information from Internet services. Additionally and through APIs, real-time or deferred information channels may be integrated into the model.

The performance of a prediction model depends largely on the quality of data, the set of algorithms, among others, but also depends on the data split configuration. Despite not having with specific and exact mechanism is fundamental to count on a strategy to establish the correct percentages of data for training, validation, and testing. Using splitting methods instead of splitting randomly to obtain the optimal configuration may be an option.

Future research must point to developing prediction models using deep learning (a combination of supervised and unsupervised learning techniques) and be focused on using data sources little used in traffic accident predictions (driver’s data and pedestrian mobility).