FORECASTING OF TRAFFIC ACCIDENTS IN ECUADOR BASED ON MULTILEVEL SINGULAR VALUE DECOMPOSITION - MSVD

Institutions and organizations need to support their planning tasks to improve decision-making, they require known information and the non-visible knowledge contained in the data. One of the main challenges for most countries is to reduce the rate of road accidents by implementing strategic actions such as scientific studies focused on supporting prevention plans. In this article, the method based on multilevel singular value decomposition (MSVD) is applied to improve the accuracy of the multi-horizon forecast of a linear MIMO-AR (multiple input-multiple output autoregressive) model. The proposed model is evaluated by means of the time series of traffic accidents in Ecuador, collected weekly between the years 2015 to 2018, the number of injured, uninjured, deceased, detained vehicles and identified drivers. The results show that the model reaches an average Nash-Sutcliffe efficiency of 89.2% for a 12-week horizon, and an average Coefficient of Determination R-2 of 98.9% for the same horizon.