期刊名称：International Journal of Advanced Computer Science and Applications(IJACSA)
印刷版ISSN：2158-107X
电子版ISSN：2156-5570
出版年度：2019
卷号：10
期号：11
DOI：10.14569/IJACSA.2019.0101166
出版社：Science and Information Society (SAI)
摘要：Around the world, road traffic accidents (RTAs) cause significant concerns for decision makers and researchers on traffic safety. The diversity, rarity, and interconnectivity of historical data on factors causing car accidents point to the need for more focused studies for analyzing, predicting, and visualizing the risk of accidents over the short and long term for preventive purposes. There are many techniques and tools applied to analyze, forecast, and visualize risk. Most RTA studies have applied linear time-series methods to forecasting the risk with limited studies applying machine-learning and deep-learning techniques, especially in Saudi Arabia. Recently, many global studies have applied long short-term memory (LSTM) networks, which can be used to automatically learn the temporal dependence structures for challenging time-series forecasting problems. This paper displays a tutorial for designing a prototype of an interactive analytical tool based on a multivariate LSTM model for time-series data to predict future car accidents, fatalities, and injuries in the Kingdom of Saudi Arabia (KSA). This interactive tool visualizes the real data with the predicted values regionally in a web browser with Python. The tutorial represents the annual data of the period between 1417 (1996) and 1433 (2013), then uses the data with some contributing factors, such as population, gender, nationality, number of vehicles, and length of road, to generate the input data and predict the future values of accidents, fatalities, and injuries up to the year 1452 (2030). After that the real and predicted values are visualized regionally on an interactive map that represents the degree of risk. Finally, the paper discusses the evaluation and utilization of the proposed prototype in the future in the field of road safety.
关键词：LSTM for time-series forecasting; deep learning; RTA; data visualization; interactive map; Saudi Arabia
