Development of lag size-based safety thresholds for skewed uncontrolled intersections

Gap/Lag acceptance is the primary basis for analysing uncontrolled intersections. Misjudgement in gap/lag acceptance imparts high risk to the drivers. The gap refers to the temporal difference between consecutive vehicles on a major road, whereas lag is a part of the available gap, occasionally coinciding with the first gap. Even though both are different in real scenarios, studies do not consider them separate. The lag acceptance behaviour of drivers must be studied thoroughly because the acceptance of shorter lags is more in developing countries due to the aggressive behaviour of drivers, which might lead to road crashes. However, such studies are very scarce compared to gap acceptance studies. A study of the lag acceptance process is essential for improved traffic safety and operational efficiency at skewed uncontrolled intersections. This study adopted machine-learning techniques to predict the lag acceptance decision of drivers to examine how it performs compared to commonly used methods. Data were collected at six intersections from various cities in Kerala, India, during peak hours. Artificial Neural Network (ANN), Logistic Regression (LR) and Support Vector Machine (SVM) models were developed, and their performance was compared. The occupancy time approach was used to determine the critical lag. The goodness of fit measures shows that the ANN model outperforms the LR and SVM models, with an accuracy of 93.6%. Furthermore, goodness-of-fit measures such as F1 score and R2 values are 0.964 and 0.892, indicating that the prediction of the ANN model is excellent. Lag sizes of less than 2.7, 3.5, and 3.0 seconds were shown to be less safe, corresponding to right-turn from major, right-turn from minor and through from major roads.