M. Govindasamy and S.K. Misra (USA)
Software Design and Development, Object-oriented Metrics, Software Testability, Bayesian Model
Software testability, one of the quality attributes, is defined as the ease of testing the software under specific input/output. A program has high testability if it tends to expose faults during testing and has low testability if it tends to hide faults even though such fault may exist in the tested code [1]. Testability is a quantifiable measure and is measured using several techniques: Domain range ratio analysis [2], Propagation, infection and execution analysis (PIE) [3], Observability and controllability analysis [4], and Neural network analysis [5]. The dynamic failure based PIE technique and the neural network analysis provide better measure for testability compared to the other two techniques in procedural programs. These techniques are cumbersome and computer intensive in object-oriented (OO) programs. In OO programs, software quality indicators, such as maintainability and reusability, are predicted using object oriented metrics [6, 7]. Current studies of quality predictors use the multivariate regression model, which poses a weak prediction model compared to the Bayesian approach. This paper defines, explores and analyzes the Bayesian model approach to predict testability using the well known validated object-oriented metrics and test metrics. This article explores the existing Bayesian approaches and evaluates the applicability of those towards predicting testability. We argue that the Bayesian Model Averaging (BMA) is a better methodological approach to developing prediction model for testability.
Important Links:
Go Back
