Before Thailand was ranked first in road accident deaths per 100,000 population in 2017, the WHO had already ranked Thailand’s road accident situation as the world’s second-worst in 2015. This ranking created the unpleasant image that travel in Thailand is unsafe. The organizations and departments involved have recognized the need to prioritize measures to decrease the worsening road safety situation, including law enforcement, education campaigns for schools, advertising media, and increasing the training hours required to obtain a new driver’s license or renew a license, as well as using technical engineering solutions. Funding for research has been provided to investigate and identify solutions to this problem in an atmosphere of economic growth focusing on travel, accident risks, and into solutions for reducing the number of accidents and deaths. Among the measures mentioned above, the National Police Bureau statistics indicated that the death rate has been decreasing; however, in 2016, the rate actually increased (Figure 2
The GDP, the number of registered vehicles, and transport sector energy consumption are likely to increase in the future (Figure 2
b–d). Therefore, in this study, we analyzed statistical data to forecast the death rate from road accidents by a time-series model using exponential smoothing, curve estimation, multiple linear regression, and path analysis, using official Thai statistical data collected over the past 20 years. The research results are as follows:
The time-series model using exponential smoothing is suitable for predicting the death rate from road accidents. Time-series techniques have also been used to forecast accidents by ARIMA [9
]. However, the Thai data set was not suitable for the ARIMA technique. The exponential smoothing technique yielded MAE and MAPE values of 1.627 and 8.1%, respectively.
Curve estimation with cubic, quadratic, and linear patterns were the three models with the highest R2 values.
Multiple regression linear model 1 found that GDP was a good economic indicator, in agreement with a previous report by Dadashova et al. [16
], and that the transport sector energy consumption level affected the death rate from road accidents, in agreement with reported results García-Ferrer et al. [12
]; the number of registered vehicles (motorcycle, cars, and trucks) had no effect.
Using multiple regression linear model 2 (where the proportion of various factors was adjusted by population), we found that the number of registered vehicles and transport sector energy consumption affected the death rate from road accidents, whereas, the other factors had no effect.
Using multiple regression linear model 3 (where the proportion of factors was adjusted for the number of registered vehicles), we found that the number of registered vehicles, number of registered trucks, and the amount of energy consumed by the transport sector affected the death rate from road accidents, whereas, the other factors had no effect.
The path analysis model showed that GDP, energy consumption, and the number of registered vehicles were factors that directly influenced the road death rate. The amount of energy consumed by the transport sector was a factor influenced by the GDP, which indirectly affected the number of road deaths.
The effectiveness of the first three models with the lowest MAPE were multiple regression linear model 3, the time-series (exponential smoothing) model, and the path analysis model, with MAPE values of 6.4%, 8.1%, and 8.4%, respectively.
When the models were used to predict the death rate from road accidents, we found that the time-series (exponential smoothing), curve estimation (quadratic), curve estimation (linear), multiple regression 2 and 3, and path analysis models forecasted decreasing fatal accident trends, which supports the data on the direction of the death rates provided by the Royal Thai Police [6
]. We found that the multiple regression linear 1 and curve estimation (cubic) models generated forecasts that contrasted the trends observed in the data provided by the National Police Bureau, whereas, the curve estimation (quadratic), multiple regression linear 1, and path analysis models predicted a zero value, and thus, are not suitable for long-term forecasting. Only the time-series (exponential smoothing), curve estimation (linear), and multiple regression linear 3 models generated predictions that were consistent with the trends present in the original statistical data.
ARIMA models have been applied to accident forecasting [9
]. However, Thailand’s data were not appropriate for applying an ARIMA model, but were suitable for exponential smoothing, and hence, could be used in forecasting. The economic growth data that were used in forecasting by applying multiple regression linear and path analysis were GDP, the energy consumption of the transport sector, and the number of registered vehicles (motorcycles, cars, and trucks).
According to our data, none of the models were found to be suitable for predicting the death rate from road accidents 10 years in advance. The road death rates will feasibly decline over the long time period of over 10 years and the various measures implemented. The economic and transportation factors considered, which reflect the economic growth of the country, had both direct and indirect effects on the road death rates. If considered in -depth, some data may be useful for informing government policy-making and for designing preventive measures to reduce the causes of accidents, especially the number of registered cars on the roads, which is directly related to the number of accidents. In addition to personal and environmental factors, the appropriate control of the rate of vehicle occupancy should be considered. The legal driving age should be reviewed, along with knowledge of traffic regulations and proven experience in safe driving, when applying for a driver’s license.
Appropriate policies are required to reduce fatal accidents in the public sector. Due to the mixed traffic road conditions in Thailand, trucks and other large vehicles share roadways with other small- or medium-sized vehicles, which may cause dangerous situations and accidents. The public sector should implement policies to rigorously control the driving speed, covering all types of vehicles and providing exclusive lanes for freight vehicles. Principally, these policies may help drive Thailand’s economic growth, consistent with the results of multiple linear regression model 1, in which positive growth of the economic factor GDP, as an overall gross product of the nation, indicated increased economic activity (e.g., import and export, and generation of jobs and income).
The public sector must create policies related to the control of the possession of vehicles, including stricter measures, such as requiring declaring a driver’s license, and consideration of traffic violation history and accident history, to possess a vehicle. This policy would be consistent with the results of multiple linear regression models 2 and 3 and the path analysis model. Energy consumption in the transportation sector was found to be connected to the recent number of registered vehicles in Thailand, which has been increasing.
Other factors affecting these policies could be investigated in terms of budgets for solving the accident problems considered in this study for mitigating and preventing accidents or deaths; for example, by providing knowledge and understanding of accidents through public relations by community leaders or organizations, or providing a driver’s license.
5. Limitations and Future Work
Model analysis involves forecasting limitations, which potentially result in the misleading prediction of trends. When attempting to forecast road accident death rates, other factors must be considered, including law enforcement measures, such as those on speed limits, drunk driving, helmet wearing, seat belt use, and phone use while driving and other distracted driving behaviors, in addition to public transport use, transport infrastructure developments, and other economic and social issues, which have yet to be analyzed systematically (lane markings, lighting, road markers, signage, intersections, warnings).
The data that were used in our analysis, due to Thailand incompletely collecting data, according to the plan that was set 20 years ago, led to the lack of many types of data in the analysis. Thailand’s accident data has several databases, which affected the consistency of the data.