Citation

Alshannaq M, Imam R, Alsarayreh DW. Int. Rev. Civil Eng. 2023; 14(5).

Copyright

(Copyright © 2023, Praise Worthy Prize)

DOI

10.15866/irece.v14i5.22730

PMID

unavailable

Abstract

The quantity and range of research examining the severity of injuries in road accidents have expanded dramatically in recent years. This research investigates the relationships between injury severity levels and their causes. It develops new effective countermeasures that may contribute to reducing the severity of injuries caused by accidents using Machine learning - Random Forest (RF) and Logistic Regression- to model injury severity of traffic accidents in Jordan through R programming language. According to Gini's Improvement Index (GII), the most significant is Accident Primary Type (APT), which has a 16% improvement index. Still, according to the P-value, all variables were insignificant. Random forest of a tree-based classification and regression methodology applied to data with 500 trees has the advantage of diminishing overfitting without affecting prediction accuracy; the number of variables tried at each split is two, and the split between them is two.

RESULTS of RF models show that the mean age of drivers was 34 years, where the highest percentage among the causes of accidents was "violation of traffic rules" with 42%, while the "abrupt/sudden lane change" was the lowest at 11%. Multinomial Logistic(MNL) Regression results found that factors such as Age, Gender, Accident Max Speed (AMS), Weekday (W), Accident Primary Type (APT), Road Surface (RS), Weather Status (WS), and Manufacturing Year (MY) are all negatively related to crash severity at all levels (property damage only (PDO), minor and major injury). In contrast, the rest of the factors, like Vehicle Registered Type (VRT), Road Lane Number (RLN), and Lighting Status (LS), are positively related at all levels. As for fatal accidents, the vehicle registered type has the highest odds ratio (36.943). Furthermore, the relative risk was substantially related to several lanes (one-lane, two-lane, three-lane) at (0.173, 5.441, 1.302), respectively.

Language: en