Cohesion prediction using information flow: an empirical feasibility study and comparison using students as inexperienced designers

Abstract

This paper describes an experimental investigation that demonstrates the feasibility of predicting weakly bound cohesive software modules [W.P. Stevens, G.J. Myers and L.L. Constantine, IBM Systems Journal (2) (1974) pp. 115–140] at the design stage of development. It compares the predictive capability of prediction systems, for weakly bound cohesive modules, based on classical regression models with systems that use binary logistic regression models. It also considers the models’ parsimony of parameters and their fit to the sampled data. Further, the external validity of the prediction systems is discussed and an appropriate mode of usage is proposed for the ‘better’ statistical models.