Prediction of pavement performance via smatrphone vibration-induced unevenness signature using machine learning.

Pavement performance evaluation plays a crucial role in guiding decisions related to the maintenance and rehabilitation of roads, ensuring their sustained optimal performance over an extended period. Contemporary smartphones come equipped with ample storage, computing power, and communication capabilities. Additionally, they feature built-in sensors that exhibit exceptional abilities in capturing information about roads and their surrounding environment. Hence, leveraging smartphones for pavement condition evaluation proves to be a worthwhile and cost-effective approach, especially for developing countries with limited pavement data and budget constraints. Several research studies were conducted regarding the mentioned constraints; they compared the smartphone vibration-based pavement condition value with an index such as the International Roughness Index (IRI). Few studies were carried out to estimate pavement condition evaluation based on other indices. This study will present a novel approach to predicting pavement condition indices measured using road surface profilers such as IRI, pavement condition index, rut depth, mean profile depth, and combination of them via Vibration-Induced Unevenness Signature (VIUS) using smartphone vibration data. Various machine learning techniques are used to obtain the best model between VIUS and the condition indices. Among these techniques, the Gradient Boosting technique outperformed the rest in terms of coefficient of determination and mean square error. Hence, the findings suggest moderate to strong correlations between certain smartphone sensor features and pavement condition indices, particularly IRI.