Travel Time Reliability Prediction by Genetic Algorithm and Machine Learning Models in Virginia

Abstract

Travel time reliability has proved to be a critical issue both in the context of traveller's choices and decisions and freight transportation. In the terminology, the temporal variability of travel time is known as reliability and is affected by numerous factors. Three of them, including traffic volume, incidents, and inclement weather, are among the most profound, and their effects have been the subject of many studies. What has made this article unique is the simultaneous implementation of a genetic algorithm with multiple machine learning methods. Also, GA could eliminate overfitting, which is a common mistake in ML models. The numerical results revealed that the performance of the prior model, KNN, enhanced significantly when GA was imposed on it. In terms of stability ratio, a 12% decrease was observed. Also, the mean squared error (MSE) for the training set and test set decreased. However, the reduction was not significant. To further illustrate the advantages of GA implementation, the number of predictions with MAPE greater than 0.05 were compared, and a notable reduction was revealed. In the final step, sensitivity analysis was done to depict how PTI responds to the fluctuations of independent variables.