Abstract:
Quantification of present pavement condition in terms of an index term
i.e., Pavement Condition Index (PCI) is one of the most important and primary steps
while taking decision related to Maintenance and Rehabilitation of Pavements. PCI
as proposed by ASTM D6433 rates pavement in seven conditions viz. Good, Satisfactory,
Fair, Poor, Very Poor, Serious and Failed. Determination of rating condition
of pavement using distress severity and extent turns out to be tedious process. Hence,
present study investigates application machine learning techniques for assessment
of present pavement condition. Three different algorithms i.e., Logistic Regression,
Naïve Bayes and K-Nearest Neighbor have been tested in the present study using
Long Term Pavement Performance database consisting of over 10,000 datapoints.
The dataset was divided into 7:3 ratio for training and testing phase. Employed
algorithms were tested based on accuracy, precision, recall and f-measure. Logistic
Regression Classifier was found to have highest accuracy of 0.92 among three
classifiers used in the study.
