Study of the interval width of features for improving forecasting efficiency of the transport flow intensity

Introduction. The complicated system of characteristic interaction between the Driver-Car-Road-Environment (DCRE) is the research subject for the scientists around the world. If we explain the patterns of phenomena present during road accidents, we will be able to model some transport processes. To do this, we need to consider a large number of characteristics, dividing them into static and dynamic. Particular attention is paid to the latter, due to the instability of the features. This category includes traffic flow intensity and weather conditions. There are a lot of methods for increasing the accuracy of predictive models, but this method has been used for the first time. Logicaland statistical validity of the selection automation of interval rages are the main feature of this method. This is necessary not only for grouping features, but also for increasing their value in a joint analysis. For example, for the intensity of traffic flows the index number can be 100 vehicles/hour (0-100, 101-200, 201-300, etc.), but it will not be effective from a prognostic point of view for the temperature interval index of 5°C (-25 - -20, -19 -15, -14 - 10,etc.). Accordingly, the goal of the work was to determine the effective forecasting of the interval’s width of traffic flow intensity (dependent feature) and weather conditions (independent features). Materials and methods. This work is a continuation of a large project on improving road traffic safety, in which similar studies have already been conducted to determine the effective interval coefficients using Spearman’s rank correlation. The values at which temperature regimes (air, soil and dew point) best describe the intensity of the traffic flow were established. For a comprehensive characterization, additional analysis was necessary to conduct of the remaining independent features. New algorithmic structures were created using the Python programming language, in which the established feature interval ranges were sequentially compared in such a way as to process all possible combinations. Each result was subjected to correlation analysis, and the probability of error was calculated. Results. As a result of the experimental selection of interval ranges, the most effective of them were determined. The selection criterion was the subsequent correlation analysis. The coefficient values greater than 0.7 or less than -0.7 were accepted. The probability of error was also calculated, and values less than 0.05 were accepted. Thus, a large number of combinations were obtained that meet the necessary conditions. Further, for each feature, the interval width was selected at which it is more often intersected with others, and in the case of the same number of intersections, it is the smallest of them. Conclusion. As a result, effective interval widths were determined in which the investigated features had been analyzed. This study in subsequent works contributed to the high-quality training of the model using the deep learning method. Thanks to this research, a program to predict the intensity of the traffic flow, depending on weather conditions with using neural networks was created.