Understanding and identifying the spatial patterns of aging traffic accident by a combined GIS and machine learning approach
With the coming of the aging society, the situation that elderly involved in Vehicle-Pedestrian (VP) collisions will occur more frequently. A large body of studies has pointed out the possible features which cause aging VP collisions while fewer studies tried to predict the potential occurrence location. This study referred to those elements and conducted with Support vector machine (SVM) selected from three machine learning algorithms (Random Forest, Neural Network, and Support Vector Machine) to create a predicted model. However, even though the important features have shown the spatial relationship and validated the previous researches, the results indicated that the output model can only predict the location where have no VP collisions occurrence. After further investigation, the present study found out reasonable explanations for the phenomenon. First, the decision of utilizing SVM was simply based on overall accuracy without considering the low accuracy in the ability to predict the location of VP collisions. Moreover, the selected features were combined with the same types which decrease the significant characteristic and increase the difficulty to discriminate. Overall, this study shows the possibility of utilizing machine learning in future relevant research. After the gap filling, the application of these techniques could potentially contribute to decreasing similar accidents in the future.