New trends in digitalization in construction have created opportunities for research and informed decision-making. Concepts like digital twins and sensorization have successfully enabled the direct collection of data from construction processes and equipment. For instance, integrating sensors into trucks transporting construction materials facilitates gathering valuable information about the equipment and the surrounding environment. This previously unattainable data can now be utilized to provide pertinent insights into the decision-making process. On one hand, accurate fuel consumption estimations are required to help optimization in construction and transportation infrastructure projects as they represent a major expense. On the other hand, despite the numerous studies conducted to detect cracks and potholes in road pavements, the classification of road types is frequently overlooked. This study aims to bridge this gap by developing a methodological framework that utilizes vibration data from sensors installed in construction trucks to predict the fuel consumption of heavy vehicles and the road category based on the pavement surface quality through which it is circulated. Given their promising results in prior research, the models Random Forest, Neural Network, and Support Vector Machine were applied to the database. The results demonstrate that vibration-based data acquisition methods combined with machine learning algorithms can accurately predict fuel consumption, identify different road categories, and can be successfully applied on a larger scale.