Learning to predict test effectiveness

The high cost of the test can be dramatically reduced, provided that coverability as an inherent feature code under is predictable. This article offers a machine learning model to predict extent which could cover class in terms new metric called Coverageability. prediction consists ensemble four regression models. samples consist vectors, where features are source metrics computed for class. labeled by Coverageability values their corresponding classes. We offer mathematical evaluate effectiveness size and coverage suite generated automatically each extend space introducing approach defining sub-metrics existing metrics. Using importance analysis on learned models, we sort order impact effectiveness. As result which, found strict cyclomatic complexity most influential metric. Our experiments with models large corpus Java projects containing about 23,000 classes demonstrate Mean Absolute Error (MAE) 0.032, Squared (MSE) 0.004, R2-score 0.855. Compared state-of-the-art our improve MAE, MSE, 5.78%, 2.84%, 20.71%, respectively.