ramos-iyer / major-and-minor-severity-accidents-a-study-of-environmental-factors-using-deep-neural-networks-and- Goto Github PK

Owing to the increase in number of vehicles and infrastructure, the need for severity analysis in road accidents has gained a fundamental importance. This research aims to create a severity detection system for major and minor accidents depending on the length of the road. In order to achieve this, the environmental factors of weather, road demographics, accident situation factors and nearby landmarks has been taken into consideration. The literature survey shows that the implementation of feature importance on a per class basis along with the model as a whole has not been widely explored. The implemented Stratified K-Fold Deep Neural Network performs better than traditional ensemble models with an accuracy of more than 89% for both, Major and Minor accidents. Along with this, to gain more information from the neural network and how it predicts, SHAP (Shapely Additive Explanations) has been applied to gain overall importance. These feature importance scores have been further explored on a per class basis using the data itself and extraction of SHAP scores for each class. The findings from this research can be used by the main four stakeholders in an accident - transportation safety planners, hospitals, medical agencies and insurance companies for proactive action and purposes.