Traditionally, offset optimization for coordinated traffic signals is based on average travel times between intersections and average traffic volumes at each intersection, without consideration of the stochastic nature of field traffic. Using the archived high-resolution traffic signal data, in this project, we developed a data-driven arterial offset optimization model that will address two well-known problems with vehicle-actuated signal coordination: the early return to green problem and the uncertain intersection queue length problem. To account for the early return to green problem, we introduce the concept of conditional distribution of the green start times for the coordinated phase. To handle the uncertainty of intersection queue length, we adopt a scenario-based approach that generates optimization results using a series of traffic-demand scenarios as the input to the offset optimization model. Both the conditional distributions of the green start times and traffic demand scenarios can be obtained from the archived high-resolution traffic signal data. Under different traffic conditions, queues formed by side-street and main-street traffic are explicitly considered in the derivation of intersection delay. The objective of this model is to minimize total delay for the main coordinated direction and at the same time it considers the performance of the opposite direction. Due to model complexity, a genetic algorithm is adopted to obtain the optimal solution. We test the performance of the optimized offsets not only in a simulated environment but also in the field. Results from both experiments show that the proposed model can reduce travel delay of coordinated direction significantly without compromising the performance of the opposite approach.