Accident Severity Detection Using Machine Learning Algorithms

Abstract

The early and correct detection of traffic incidents can effectively reduce personal casualties and property damage, and improve the effectiveness of macro-control and scientific decision-making of traffic. Traffic incident data unbalance plays an important role in the impact of detection. Therefore, the traffic incident detection method based on factor analysis and weighted random forest (FA-WRF) is proposed. By processing the traffic flow parameter change rule to build the original incident variable. Factor analysis (FA) method is used to reduce the dimension of the original incident variables. Bootstrap improved algorithm is used to pre-determine the training set data extraction standard. The MCC coefficient value is calculated for the classification effect of the decision tree during training, and is used for each tree as a weight value, such that the trees with more powerful classification ability have higher voting power in the voting process, thus improve the overall classification effect of the random forest (RF) algorithm for unbalanced data. The detection performance is evaluated by the following universal parameters such as detection rate, false alarm rate, classification rate and receiver operating characteristic area under the curve. The experimental results indicate that the FA-WRF-based model has the best classification result. Meanwhile, it is comparable to Support Vector Machine in handling unbalanced data classification.