Embodiments of the present invention provide an approach for adapting an information extraction middleware for a clustered computing environment (e.g., a cloud environment) by creating and managing a set of statistical models generated from performance statistics of operating devices within the clustered computing environment. This approach takes into account the required accuracy in modeling, including computation cost of modeling, to pick the best modeling solution at a given point in time. When higher accuracy is desired (e.g., nearing workload saturation), the approach adapts to use an appropriate modeling algorithm. Adapting statistical models to the data characteristics ensures optimal accuracy with minimal computation time and resources for modeling. This approach provides intelligent selective refinement of models using accuracy-based and operating probability-based triggers to optimize the clustered computing environment, i.e., maximize accuracy and minimize computation time.