Application of Data Augmentation Techniques in Predicting Travel Time Reliability: Evidence from England

This study investigates the effectiveness of data augmentation techniques like noise creation, scaling, shifting, and Grey models (GMs) for improving prediction model performance, especially in scenarios with small datasets. Despite concerns about GMs’ suitability for data augmentation, the study finds that these techniques can help mitigate the overfitting issue often encountered with small datasets. The research uses machine learning models to evaluate the performance of these augmentation techniques. To do so, the authors utilized data on the average speed, delay, and travel time reliability of strategic road network (SRN) in England for 72 successive months, from April 2015 to March 2021 on which were calculated on the network level. The study applies augmentation techniques to expand the dataset and evaluates their performance using various machine learning models. It concludes that Grey models (GMs) can be reliable tools for data augmentation, especially for simple problems. However, it highlights potential issues such as over-augmentation, inappropriate augmentation, and poor-quality augmentation. The study underscores that GMs offer logical and acceptable numerical results and are easy to implement, dispelling earlier doubts about their effectiveness. To the best of the authors’ knowledge, methodologies have been proposed by researchers in the field of data augmentation techniques for numerical data, but none of them have focused on utilizing GM models as a solution for augmenting numerical data.