Classification of pavement climatic regions through unsupervised and supervised machine learnings

Abstract This study extracted 16 climatic data variables including annual temperature, freeze thaw, precipitation, and snowfall conditions from the Long-term Pavement Performance (LTPP) program database to evaluate regionalization for pavement infrastructure. The effect significance of climate change were firstly evaluated using time as only predictor t-test. It was found that both temperature humidity increased in most States. Around one third 800 weather stations record variation precipitation classifications a few them show significant over based on results logistic regression analyses. Three unsupervised machine learning Principle Component Analysis (PCA), factor analysis cluster conducted identify main component common factors variables, then classify datasets into different groups. Then, two supervised methods Fisher’s discriminant Artificial Neural Networks (ANN) adopted predict regions data. Results PCA are first principle components factors, accounting 71.6% variance. 4-means clusters include wet no freeze, dry snow freeze. best k-mean clustering suggested 9 with more clusters. Both linear ANN can effectively multiple variables. performs better higher R square low misclassification rate, especially those layers nodes.