Prediction of Pavement Condition Index from Visual Surface Condition Rating Using Regression Analysis

Pavement performance evaluation is an essential part of road asset management, enabling the planning of budgets and maintenance priorities. Road agencies all over the world have to deal with the challenge of accurately and reliably conducting pavement condition assessments for their networks as a first part of the evaluation process in the face of scarce financial resources. In Kenya, the Kenya Roads Board (KRB) has applied a visual surface condition rating methodology based on observing at least two pavement distress types to quickly assess the country’s road network at a manageable cost. Additionally, for a portion of the network constituting around 66.30% of the total network (9431 km of flexible pavements), pavement condition indices (PCIs) have been determined through the more comprehensive ASTM methodology that involves a detailed evaluation of observed pavement distresses. However, to enhance planning efficiency and improve the existing maintenance management regime based on more detailed observation, it is desirable to have complete PCI data for the entire road network. This paper applied regression analysis to predict and impute the missing PCI data by utilizing available data. Among five regression modeling techniques (multiple linear, polynomial, decision tree, random forest, and support vector), support vector regression was found to give the best fit to the data, with R², RMSE, and MAE values of 0.47, 13.76 and 9.23 respectively. These values, however, only moderately explain the variance in the data, and there is a need to explore a more advanced imputation methodology going forward to accurately impute the missing PCI values and improve the quality of PMMS.