Abstract

The descriptive analysis could be used to find the trends and patterns in historical data. In this project, we will use descriptive analysis to describe the car accidents in Washington, D.C., between 2009 to 2020. The dataset was downloaded from the District Department of Transportation (DDOT), the department responsible for car accidents in Washington, D.C. We have applied multiple analytics and statistical models to find the relationships between different variables and the patterns and trends among the data, such as correlation analysis, confidence interval, One-Way-ANOVA, decision tree, and visualizations. The project aims to find the common reasons for the accident and help DDOT find ways to reduce and eliminate the accidents in the area. The statistical and analytical tools examine multiple features to find the patterns and trends among the datasets. Four main findings were found after analyzing the data. First, the main reason for most crashes is drunk people, either drivers or pedestrians. The second finding is that the top reason that causes deadly accidents is speed. Also, we have found that most of the accidents are safe. In addition, we found the top ten streets with the highest number of accidents, and we found that they are located on the north side of the town.

 

ResearchGete

The project was publised in International Journal of Advanced Computer Science and Applications (2023)

 

Introduction

What are the reasons that car accidents are one of the most dangerous events on roads? World Health Organization says that around 1.3 million persons die because of car accidents and cost around 3% of most nations’ gross domestic (2021). According to the National Highway Traffic Safety Administration of the United States Department of Transportation, this is the highest six-month rise ever recorded in the Fatality Analysis Reporting System’s history. In the first half of 2021, a projected 20,160 individuals died in car accidents, rising 18.4 percent over the same period in 2020. Since 2006, this has been the highest number of expected fatalities in that period (National Highway Traffic Safety Administration, 2021). Many two kinds of reasons that lead to crashes on roads. First, external effects are the effects that the driver cannot avoid or is hard to avoid, such as the weather conditions and road situations. Second, internal effects are related to the car’s driver, such as the driver’s health condition, distractions, or car’s tier issues. According to Templeton Smithee Hayes Heinrich & Russell, LLP (2020), Driver distraction is the leading cause of automobile accidents. Intoxicated drivers, speeding, hostile conduct, rain, failure to obey traffic signs, night driving, vehicle troubles, tailgating, and improper turns and driving lean are all factors to consider.

According to data acquired from the District Department of Transportation, about 258,000 accidents in Washington, D.C., between 2009 and 2020. (District Department of Transportation, 2021). Because it is the capital of the United States, Washington D.C. is one of the most important locations in the country. It includes all government offices, tourist attractions, and educational institutions. Furthermore, according to demographic data, the population in 2020 will be 689,545 people living in 68.34 square miles (Data Commons, n.d.).

This project explores and performs a descriptive-analytical analysis of the car accidents done in Washington, D.C, between 2009 and 2020. In order to find insights and patterns in those accidents and understand the reasons and the relationships between different variables that lead to those crashes.

Objective

This project will analyze the car accidents in the Washington, D.C area from 2009 to 2020. The researcher will examine the common factors between accidents, the locations of the accidents, the car crashes factors that may cause deaths or injuries compared to the number of accidents. Also, find the factors that significantly correlate with the number of injuries and accident elements. The accident elements are any vehicle involved in an accident. In addition, to rank the accidents into groups.

Finding patterns in many incidents gives a clear view of the likely causes that lead to an accident. It will also reveal whether any issues need to be addressed to limit the number of incidents. Because automobile accidents result in numerous injuries, the causes of such injuries will be investigated. On the other hand, many accidents result in merely automobile damage and no human injuries. In addition, assessing the automobile collision location will offer helpful information about areas where authorities should focus their efforts. It can also determine which areas have a high number of injuries or accidents that result in fatalities.

Dataset

Original Dataset

District Department of Transportation (DDOT) provides a high-quality dataset containing the accidents in Washington D.C. The data were collected by DDOT and Metropolitan Police Department (MPD). The data were downloaded contains 258,122 records and 60 features. However, 19 features will be dropped because it has no relation to the kind of analysis performed on the data. Also, all car accidents that happened before 2009 were dropped. After dropping the unrelated column and data before 2009, those accidents have no vital data. There are seven columns added to the datasets. Because the dataset is very detailed, those columns were added to aggregate some columns to calculate the number of injuries per accident.

Preprocessing And Explore the Dataset

The shape of the final dataset that will be analyzed contains 237,193 and 55 columns. The following table shows the features with a brief description for each column (District Department of Transportation, 2021).

 

Explore the Data

To understand the data and the distribution, we need to present statistical information and illustrate visualizations.

First, using describe function on the data frame that contains the data provide vital information about the data. The table below shows the count, mean, standard deviation, min, max, 25%, 50%, and 75% for the added columns.

	TOTAL FATAL	TOTAL MAJORINJURIES	TOTAL UNKNOWNINJURIES	TOTAL ACCEIDENTELEMENTS	TOTAL INJURIES	TOTAL INJURIES ELEMENTS
count	237193	237193	237193	237193	237193	237193
Sum	476	25,980	18,161	532,660	114,316	646,976
mean	0.002007	0.109531	0.076566	2.245682	0.481954	2.727635
std	0.045961	0.459874	0.311264	0.66367	0.838257	1.096224
min	0	0	0	0	0	1
25%	0	0	0	2	0	2
50%	0	0	0	2	0	3
75%	0	0	0	3	1	3
max	2	51	16	17	51	55

Table: New columns description

The previous table shows the data distribution. By reading the number, we can find much helpful information that makes understanding the data more straightforward. For example, it can be noticed that the maximum number of fatal is two, which is a good indicator that although the number of accidents is high, the fatal cases are too few.

Also, visualization is considered one of the best ways to explore and understand the data. There are some charts generated to understand the data range and distribution.

 

This chart in Figure 1 shows the total number of elements and injuries based on the sector. It can be noticed that there are some outliers and the car crashes distribution based on these regions.

These box plots show the relationships and distribution between the total number of injuries and elements, total injuries, total accident elements, and total major injuries based on the accident category rank.

 

The System

This project examined many statistical and analytical models to find the patterns and trends in data. The first model that will be used is the correlations between features. This model can provide a good idea about how columns interact with each other. If there are high correlations this indicated, we can analyze these columns and find if there is an actual relationship or not. As known, correlation does not mean causation, but this phrase has been wrong in some cases. Also, a decision tree analysis is performed to find the factors that lead to fatal accidents. To understand in which cases the accidents could have any fatal people. In addition, visualizations are a great way to explore the data and find the outliers and data distribution. Understanding the data make the analyzing process done in the right way. If data were understood well and in context, this would help to understand the result better. In addition, some statistical models could be performed to test if there is a significant relation between features, such as Spearman correlation and ANOVA analysis.

The statistical and analytical tests that will be performed can answer many questions. For example, is there a relationship between the location and the number of accidents or injuries? Is there a relationship between the weak days or weekend days and the number of accidents or injuries? What are the common factors between accidents cause deaths, major or minor cases? These are some questions this search will answer using data analytics tools. In addition, answering those questions contributes positively to identifying the car accident problem and finding solutions to reduce the number of accidents.

On the other hand, using visualizations will make understanding the data much more straightforward by visualizing the data. It makes the data complexity present in a way many people can understand. Also, charts and maps can visually answer several questions. Because we have geospatial data, we can present the accident on maps, quickly understanding the locations that hold a high or a low number of accidents. Also, we can find the locations of the significant injuries or where precisely the taxis had accidents. Besides, a timeline chart, which is an excellent way to find the number of incidents within a time range, can provide many answers to understand the issue.

System architecture

In the analyzing phase, many steps are performed to get the analyzed results and find the patterns and the relationships between different variables.

First, to prepare the data for some analysis, we need to add a column to indicate if the accident is fatal. This will help to perform the decision tree analysis.

Second, convert string and categorical data into numbers which makes applying statistical model applicable. All statistical models cannot work with non-numerical data. Five columns were converted from string to number: ADDRESS, NEARESTINTSTREETNAME, NEARESTINTROUTEID, INTAPPROACHDIRECTION, and Rate.

Finally, we need to group the data to apply the statistical models. In this project, the data were grouped by accident rates.

Data Analytics Algorithms

The data visualizations were done in this project show the relationships and trends among the datasets. Most of the charts were generated by Tableau software, and the charts were implemented after the data got cleaned and prepared for analysis.

Also, some data analytics algorithms are used to prepare the dataset for the statistical and analytical models. The first algorithm was used is the cut function for binning (Jain, 2020). This function allows us to classify the accidents using the total accident elements and injuries, making the analytics operations more resealable. In our case, we apply the statistical models to five groups.

The second algorithm was used to label the categorical variables. Because the statistical models cannot understand the string data types, we need a way to convert the string values into numerical values.

Data Analytics and Statistical Models

Seven statistical and analytical methods were applied to the data to understand the relationships between the different variables and answer the questions we asked in the introduction part. Some of these tools are descriptive, inferential, and advanced tools.

Descriptive Models:

 The first model was used is confidence intervals. Confidence intervals give the estimated value in a variable to have happened with 90% and 95% probability. This model was examined multiple variables to understand that the most number might appear in most cases. For example, what are the total injuries and accident elements that could happen in 90% of the accidents accrued in Washington D.C, and what are the total injuries in 95% of the car crashes?

Then, we performed the correlation analysis to find the relationship between the different features. There are 15 features were used in this analysis: total vehicles, total pedestrians, pedestrians impaired, drivers impaired, total taxis, total government, speeding involved, fatal passenger, total fatal, total major injuries, total minor injuries, total unknown injuries, total accident elements, total injuries, and total injuries and elements.

Inferential Models

The third statistical model was used is ANOVA to examine an independent variable with two dependent variables.

The fourth, MANOVA which allows us to examine an independent variable with more than two dependent variables.

Advanced Models

The fifth one is the decision tree. We used decision tree analysis to find the reasons that lead to fatal accidents.

Visualizations

Finally, to see the model results, it needs to visualize them into charts, making it easy to understand the patterns and identify any relationships. Data visualizations could present patterns and trends in the dataset that are hard to find by looking at the values shown in the data, especially if the dataset was relatively large. Many types of visualizations could be used to illustrate the data. For example, the bar chart shows the frequency of the categorical variables. The map shows the geospatial points on a map, which helps find helpful information that could be used to find the car crashes patterns.

Experimental results and analysis

 Confidence Interval

First, we perform the confidence intervals on multiple features: total injuries and elements, total injuries, total accident elements, total fatal, total major injuries, and total minor injuries. This can give us the chance of total injuries and vehicles in 90%-95% in the accidents. The results as follow:

Confidence Interval based on rates

Rate	count	TOTAL_INJURIES_ELEMENTS		TOTAL_INJURIES		TOTAL_ACCEIDENTELEMENTS		TOTAL_FATAL
		ci95_hi	ci95_lo	ci95_hi	ci95_lo	ci95_hi	ci95_lo	ci95_hi	ci95_lo
extreme	6	50.76	35.56	47.96	33.03	3.32	2.01	0	0
high	49	17.94	15.97	12.46	9.20	7.34	4.89	0.14	-0.02
medium	129	11.84	11.58	7.08	6.15	5.55	4.63	0.07	0.005
low	121254	3.48	3.47	0.90	0.89	2.58	2.58	0.001	0.001
lowest	115755	1.91	1.91	0.03	0.03	1.88	1.88	0.002	0.001

Table 6: Confidence Interval based on accident rates

Confidence Interval based on variables

Variable	90%		95%
	Low	High	Low	High
Total injuries and elements	2.723932	2.731337	2.723223	2.732046
Total injuries	0.479122	0.484784	0.478580	0.485326
Total accident elements	2.243440	2.247923	2.243010	2.248352
Total fatal	0.001851	0.002162	0.001821	0.002191
Total major injuries	0.107977	0.111084	0.107680	0.111381
Total minor injuries	0.293638	0.298074	0.293213	0.298498­

Table 7: Confidence Interval based on variables

Correlation Analysis

The second model that was performed is the correlation analysis. The correlation analysis helps find the relationships between different variables to understand how the data related to each other and what factors have come together. There are two kinds of correlations: the positive correlation between 0.5 to 1.0 and the negative correlation between -0.5 to -1.0. The more significant number indicates to high correlation while the smaller number indicates weak correlations.

This visualization shows the correlation between the different variables. The correlation between 0.5 to 1.0 indicates a high positive correlation. On the contrary, the correlation between -0.5 to -1.0 indicates to high negative correlation. The chart gives vital information about how the feature relates to each other. However, if we ignore the similar features such as the total vehicles and total taxis because all of them are cars, it is normal to see a high correlation between these columns. The exciting result could be seen from the columns that have no relationships. For example, the total injuries and total taxis have 0.88, which indicates a high positive correlation. We can say that as the number of taxis accident accrue affects the number of injuries positively. In other words, taxis accidents cause more injuries than other vehicles involved in an accident. Nonetheless, we can conclude from this chart that there are high and low significant relationships between the different variables as follow:

There are many significant relations between different variables, here are the most notable once:

• the total number of injuries and accident elements with taxis.
• the total number of injuries and speed and taxis.
• the speed and
  • total accident and injuries
  • total injuries
  • total major injuries
  • total taxies
• major injuries with taxies
• Total number of accident elements with
  • fatal passenger
  • minor injuries
  • total fata
  • pedestrians
  • government
  • driver impaired
• Total number of vehicles involved in the accidents with:
  • Fatal passenger
  • minor injuries
  • total fatal

One-Way-ANOVA

The third statistical model that was applied is the One-Way-ANOVA algorithm tests the differences between one independent variable and two dependent variables. This can help find if the data are random or there is a significant relationship between the independent and dependent variables. Here are the results that we found:

Independent variable	Dependent variable 1	Dependent variable 2	p-value
TOTAL_INJURIES_ELEMENTS	DRIVERSIMPAIRED	SPEEDING_INVOLVED	0.03978
TOTAL_INJURIES_ELEMENTS	TOTAL_FATAL	SPEEDING_INVOLVED	0.03974
TOTAL_ACCEIDENTELEMENTS	DRIVERSIMPAIRED	PEDESTRIANSIMPAIRED	0.000154
TOTAL_FATAL	TOTAL_TAXIS	DRIVERSIMPAIRED	0.01528
TOTAL_FATAL	TOTAL_GOVERNMENT	DRIVERSIMPAIRED	0.05858
TOTAL_FATAL	TOTAL_TAXIS	SPEEDING_INVOLVED	0.04698
TOTAL_FATAL	TOTAL_GOVERNMENT	SPEEDING_INVOLVED	0.11406
FATALPASSENGER	TOTAL_TAXIS	SPEEDING_INVOLVED	0.03778
TOTAL_MAJORINJURIES	PEDESTRIANSIMPAIRED	SPEEDING_INVOLVED	0.28874

Table 8: One-Way-ANOVA results

The table above shows significant relationships between the independent and dependent variables where the p-value is less than 0.05. It can be seen that the total injures and elements have a significant relationship between impaired drivers and speed. Also, the total injuries have a significant relationship with the total fatal and speed. However, the total accident elements variable has a significant relationship with drivers impaired and pedestrians impaired. In addition, it can be noticed that the total fatal has significant relationships between multiple dependent variables: total taxis & drivers impaired, total taxis & speed. Nevertheless, there is no significant relationship between total government & drivers impaired and total government and speed.

MANOVA

Fifth, MANOVA, this model is like the ANOVA. Nevertheless, the difference is examining more than one dependent variable to fit MANOVA, which is in our project is RATE column because it contains five groups. The following images show the results for the following variables:

• The first one shows Rate and:
  • TOTAL_INJURIES_ELEMENTS
  • TOTAL_ACCEIDENTELEMENTS
  • TOTAL_MAJORINJURIES
  • TOTAL_MINORINJURIES
  • TOTAL_FATAL
• The second image shows the Rate and:
  • DRIVERSIMPAIRED
  • SPEEDING_INVOLVED
  • TOTAL_FATAL
  • TOTAL_TAXIS
  • TOTAL_GOVERNMENT

 

 

The sixth model used is a kind of machine learning model, a decision tree algorithm. The decision tree can help find the factors that lead to a specific event. In this project, we use a decision tree to find the factors that lead to fatal accidents, which could help understand the causes that might lead to deadly accidents

This decision tree shows that speed is a significant cause of deadly accidents. This chart shows four-level depth, which gives a scenario if the accident has the speeding case and the total injuries and accident elements less than 6.5 and there are government cars involved. The chance the accident has a fatal case is high. The model accuracy is 99.8% which is high accuracy.

Data Visualization

As known, data visualizations are a valuable way to understand the data. All charts in this section were generated by Tableau software (Chabot et al., 2021), which makes creating the chart straightforward and provides a feature that allows the developer to create attractive visualizations. This part contains four charts describing the data distribution to find the trends and valuable information. The first chart shows the car accidents over time, and the second shows the crashes over time but not the lowest and low ones. The third visualization is bar chart demonstrations to top 20 streets hold the most accidents more than others. The final one shows the map of the District of Columbia with the top 10 streets and all the accidents show in points.

Total number of accidents over time

This chart shows the accident number from 2009 to 2020. It can be seen that the number of accidents rose from 2009 to 2016 from 11,982 to 26,470 respectively. After, the numbers changed slightly from 2016 until 2019. However, the accident number hit the peak in 2019 with 26,711 accidents. In contrast, car crashes dropped significantly in 2020, with 18,230 accidents. The reason for that drop because of the Covid-19 lockdown.

 

Top 20 streets with the number of accidents

This bar graph represents the top 20 streets with the number of accidents in those streets sorted ascending. It can be noticed that most accidents happen in the northwest regions. From this chart, we can conclude that the streets located in the northwest have the highest chance of having accidents more than the other regions in Washington D.C.

Top 10 streets that contained accidents

The map shows the top 10 streets that contained accidents between 2009 and 2020. It can be seen from this map that the accidents focused on the city centre and the whole streets that lead out of town from the northeast side. However there are some accidents on both sides (northeast and southwest), but they are much fewer than those on the northwest side. This diagram helps find the streets with the highest number of accidents to study the reasons and find ways to solve them.

 

Total number of extreme, high, and medium accidents

This chart shows the number of extreme, high, and medium accidents from 2009 to 2020 divided by quarters of each year. The number shown in the bars represents the total number of injuries and the number of vehicles involved in those accidents. It can be seen that the extreme accidents stopped in 2013. Also, the number of these accidents hit the peak in 2012 with around 163 accidents with 144 in the total injuries and vehicles in the second quarter. Also, the highest number of extreme accidents happened in 2009 in the first quarter, with 90 accidents. This chart helps to find accidents distribution through the time series and understand the relationship between the kind of accident and the year those accidents happened.

 

Conclusions

In conclusion, this project studies the dataset related to car accidents in Washington, D.C., between 2009 to 2020. The data has multiple processes to be ready for analytical models. First, we cleaned the data by dropping the unrelated columns and null values that the known ways could not fill. Then, the data were explored to understand the data distribution and find the columns that may benefit this analysis. After, the data was prepared to be ready for the analytical and statistical models. The preparation processes include classifying the accidents into five groups (lowest, low, medium, high, extreme) and creating a separate dataset that contains the categorical data mapped the string values into numbers. Then, we applied correlation analysis, ANOVA, confidence interval, decision tree model, and visualizations. These models were used to find the trends and patterns among the data and find any significant relationships between different variables. This project aims to help the authorities to understand the car accidents within the area so they can find the proper solutions to reduce the number of accidents and fix any issues that can be fixed. We have found that most crashes in the area are caused by impaired persons, while deadly accidents happen if one of the accident cars is driving fast or a government car is involved in the accidents. Also, we have found a significant relationship between the number of deaths and the number of taxis involved in the accidents. In addition, most roads that hold the accidents are located in the north area of the district.

On the other hand, this project could be enhanced by analyzing the crash address and zip codes. There are some issues in the address feature that need to be handled. For instance, some addresses need to be completed, and others contain missing numbers or street names, to name a few. Also, we have found that there needs to be more information about impaired people before 2015. This issue needs to find the proper way to solve. If these problems are fixed, the results could be better.