Abstract:
Pedestrian safety is a growing concern, especially in developing nations with
heterogeneous traffic conditions. One of the common safety issues at signalized intersections
is pedestrian-vehicle conflicts, as they can occasionally result in serious collisions. The
investigation was carried out at four signalized intersections in Kollam, Kerala. The study
focused on conflicts caused by pedestrian signal violations and permissive left-turn
movements. The study used surrogate safety measures such as post encroachment time (PET),
time to vehicle (TTV), time to accident (TTA), and deceleration to safety time (DST) to analyze
conflicts between pedestrians and automobiles fastly and proactively. The conflicts were
divided into highly severe, severe and normal conflicts using k-means clustering. A support
vector machine (SVM) model was developed to predict the severity level of conflicts. The
study revealed that pedestrians' signal-violating behaviour caused 57% of the highly severe
conflicts at signalized intersections. The study proposed threshold values of the conflict
indicators for each severity level. The developed severity prediction model achieved 97%
accuracy in predicting the conflicts' severity level. The study's findings aid in evaluating
pedestrians' safety at signalized intersections
