Industrial Automatic Control Systems and Controllers Annotation << Back Neural Network Methods and Models for Predicting Road Accidents and Reducing Their Risks Tyutyunnik V.M., Uthaib Masar Abed Based on the study of more than 50 publications, methods and models that are commonly used by authors on different continents for predicting road accidents and developing technologies to reduce their risk are identifi ed and summarized. Prediction results are analyzed on the basis of two-dimensional convolutional (convolutional) neural network models, longterm short-term memory, optimization models, open-source, recurrent blocks, particle swarm, graph neural network, machine learning decision tree classifi cation, statistical model, etc., as well as using different methods and indicators, as well as using various methods and indicators: processing time, processing accuracy, data accuracy and completeness, F1-estimation, area under the ROC curve, reproduction accuracy curve, Cohen's kappa statistic, mean absolute error. Seven types of shortcomings of neural network methods and models of traffi c accident prediction are identifi ed, which do not reduce the intensity of research both in the direction of applying different models and methods and in creating new ones. The assumption is made that in the near future there will be special maps of forecasts of road accident probabilities for all roads and intersections of all countries of the world, by analogy with the currently existing geographical maps and electronic navigators. Keywords: road accidents, risks, forecasting, methods, models. DOI: 10.25791/asu.12.2024.1553 Pp. 47-56.