A study of the applicability of complexity measures

摘要: A study of the predictive value a variety syntax-based problem complexity measures is reported. Experimentation with variants chunk-oriented showed that one should judiciously select measurable software attributes as proper indicators what wishes to predict, rather than hoping for single, all-purpose measure. The authors have shown it possible particular or other factors serve good predictors some properties program but not others. For example, predictor construction time will necessarily correlate well number error occurrences. M.H. Halstead's (1977) efforts measure (E) was found be better two nonchunk evaluated, namely, T.J. McCabe's (1976) V(G) and lines code, at least chunk predicted E in every case. >