Data Analysis in Pavement Engineering: An Overview

Extensive studies on data analysis have been conducted to address pavement engineering problems including material and structure design, performance evaluation, maintenance, and preservation. This paper summarized and discussed more than 40 types of data analysis methods including statistical tests, experimental design, regressions, count data model, survival analysis, stochastic process models, supervised learnings, unsupervised learnings, reinforcement learnings, and Bayesian analysis applied in pavement engineering. Generally, traditional statistical regression models are proper for significant factors quantification and pavement performance predictions with explicit model equations and meanings of parameters. The supervised machine learnings are powerful in prediction, dealing with large data volume or unstructured data such as pavement distress images, sounds, and other unprocessed signals. The unsupervised machine learnings are usually used to pre-process data by reducing the dimensionality, extracting common factors of variables, and clustering the data samples. Selecting proper models and their combinations will be the key for the increasing accumulation of historical pavement performance data, as well as the big data from automatic pavement evaluations and pavement instrumentation in future practices and studies.