Improving the Accuracy of Early Software Size Estimation Using Analysis-to-Design Adjustment Factors (ADAFs) and Non-Functional Requirements (NFRs)

Abstract

Estimation of software size, development effort, and cost are essential for project planning, which, in turn, is an important component of software project management. Inaccurate early software size estimation (overestimation or underestimation) leads to incorrect effort and cost estimation of a software project which, in turn, may lead to project failure. This inaccuracy in early estimates of software size is mainly because of the limited information available early in the Software Development Life Cycle (SDLC). Another source of inaccuracy is the failure to incorporate information about Non-Functional Requirements (NFRs) during early software size estimation. Adequately addressing these two problems is, therefore, expected to yield tangible benefits for software project managers, especially during early project planning activities. This thesis introduces a new class of metrics called Analysis-to-Design Adjustment Factors (ADAFs) to quantitatively capture the impact on early software size estimation of additional information introduced especially when transitioning from the analysis phase to the design phase by comparing the Analysis Class Diagram (ACD) and the Design Class Diagram (DCD). ADAFs are calculated for four different class diagram metrics – Number of Classes (NOC), Number of Attributes (NOA), Number of Methods (NOM), and Number of Relationships (NOR). The applicability of these ADAFs is evaluated through practical, theoretical, and empirical validation methods. Moreover, the utility of these ADAFs is assessed by comparing the accuracy of two existing early software size estimation models before and after the application of ADAFs. Results of this comparison indicate that after the application of ADAFs the prediction accuracy is significantly improved for Model 1 (i.e. on average, 34% reduction in Mean Absolute Residual (MAR) and 43% reduction in Median Magnitude of Relative Error (MMRE)) as well as for Model 2 (i.e. on average, 49% reduction in MAR and 59% reduction in MMRE). Regression-based models employing problem domain metrics have also been built to predict these ADAFs. All of these models are statistically significant (p-values < 0.05) with R2 values between 0.42 and 0.88. Furthermore, domain/category-specific regression-based early software size estimation models using ADAF-adjusted ACD metrics have also been built and validated. A comparison of prediction accuracy of these models with the Objective Class Points (OCP)-based model indicates a significant reduction in errors (i.e. on average, 16% reduction in MAR and 24% reduction in Mean Squared Error (MSE)). Apart from looking at ADAFs, we have also investigated the impact of the security-based NFRs related to the data entry validations on early software size estimation. For this purpose, we used Software Non-Functional Assessment Process (SNAP) to calculate SNAP Points for Data Entry Validations (SPDEV). We constructed and validated a regression-based early software size estimation model using SPDEV and ADAF- x adjusted ACD metrics. A comparison of prediction accuracy shows that this model has achieved significantly better results with respect to the ADAF-adjusted ACD metrics-based model (i.e. 21% reduction in MAR and 25% reduction in MSE) as well as the OCP-based model (i.e. 28% reduction in MAR and 43% reduction in MSE). Overall, the methods we have proposed for improving the accuracy of early software size estimation have shown promising results. Therefore, utilization of these methods is expected to be beneficial for project managers in planning for their software projects.