Calibrating Probabilistic GUI Testing Models Based on Experiments and Survival Analysis

Models abstract reality. Although abstract, such models can capture the essence of real world phenomena as long as they are sufficiently accurate. The development of new techniques or its usage in a different environment cannot always be satisfactory and can be expensive. In this paper we propose a novel strategy to create GUI probabilistic testing models and calibrating them with the aid of real experiments based on survival analysis. Survival analysis is used to transform exact responses of the real experiment into probabilistic predictions, comparable to the responses obtained from our testing models. Thus calibration means searching for model parameter instances that yield model predictions almost equal to survival analysis predictions. Using our strategy, we improved the accuracy of our models showing that the models has a result closely with a real experiment.