Identification of Road Traffic Accident Black Spot Locations and Accident Predictions Using Artificial Neural Network (A Case Study in Adama City, Ethiopia)

Abstract

Globally a total of 1.35 million people die annually due to road traffic crashes; in the context of an estimated 20 to 50 million serious injuries sustained in crashes around the world every year, Road traffic accident deaths in Ethiopia reach 29386 or 4.81% of total death. In Adama city road traffic accidents that happened from 2016-2020 were; fatal (196), severe injuries(360), slight injuries(480), property damage only (960), and total cost damage of 32,707,538ETB happened in Adama city in five years, One of the major problems related to road injuries and deaths were the presence of black spots. This study aimed to identify the traffic accident black spot locations and to develop an ANN model which use to predict traffic accident numbers in accident point weightage per five years. Accident data were collected from the Adama city police office and traffic volume, road geometric parameters, and spot speed was collected from selected road sections during the site investigation. The purposive and judgmental sampling method was used to select the top three accident concentrated streets in Adama city. The three targeted streets were Derartu roundabout to Asella outlet, Derartu roundabout to Harar outlet, and Derartu roundabout to Wonji outlet, In this study, the dependent variables were accident numbers in accident point weightage (APW) per five years. The independent variables were Spot speed (50th and 85th percentile speed), average annually daily traffic volume (AADT), and road geometric parameters (traffic lane width, pedestrian lane width, number of access, median width, and road layout). Accident data collected from the Adama city police database were analyzed into point weightage approach, the threshold value 46.7APW/5years were used to identify top seventeen (17) black spot locations out of forty-five (45) samples which have APW/5years values greater than 46.7, collected traffic data from selected road section were analyzed into AADT, collected speed were analyzed into cumulative percentile speed to identify the 50th and 85th percentile spot speed. Ten (10) ANN model were developed and out of them, model number-5 was selected as the best model for accident predictions. The R2 -value was used to choose the best-fit model. The highest R2 -value was obtained for the ANN around 0.97502, demonstrating that the ANN provided the best prediction relative to the predicted and target outputs. The models analytical equation were APWn = b2 + LW2*log sig (b1 + LW1 * Xn). Training the developed model with existing data and next five years traffic accidents were predicted using this model. MATLAB2016 software was used for developing the ANN prediction model. The problem observed on more blackspot road sections was: Over speed, poor or no message Sign, deficiency of road marking, narrow walkway width, and narrow traffic lane width. Suggested Countermeasures were: Speed calming measure (speed break and traffic law enforcement), Provision of warning marking and pedestrians cross marking, widening traffic lane and walkway width, filling potholes and disparate road surface, in addition, to reduce traffic accident on the main road allowing the Bajaj and Taxi to use the secondary road is recommended. The current prediction doesn’t consider the driver's characteristics (behaviors), road surface conditions and pedestrian volume, etc. Future work might focus on how to improve the prediction performance of ANN models by incorporating these parameters as explanatory variables.