Effort estimation for corrective software maintenance

Abstract

%675$&7This paper reports on an empirical study aiming at constructing cost estimation models for corrective maintenance projects. Data available were collected from five maintenance projects currently carried out by a large software enterprise. The resulting models, constructed using multivariate linear regression techniques, allow to estimate the costs of a project conducted according to the adopted maintenance processes. Model performances on future observations were achieved by taking into account different corrective maintenance task typologies, each affecting the effort in a different way, and assessed by means of a cross validation which guarantees a nearly unbiased estimate of the prediction error. The constructed models are currently adopted by the subject company.

A Slice-Based Estimation Approach for Maintenance Effort

“…For example, previous studies [2,3,10,16], determine the source-code change between two consecutive versions either from CVS logs, using some computer aided software (CASE) tools, or system utilities such as diff. When source-code changes are submitted using the Software Configuration Management (SCM) tools (e.g., Subversion, CVS, and ClearCase) best practice is for developers to commit a brief explanation of the change into the change log, which is saved collectively with the source-code deltas in the SCM repository.…”

“…The suggested model estimates the efforts that are related to both the functionality updating and anti-regressive activities. The maintenance-effort estimation that involves the convention of linear regression analysis was introduced by De Lucia et al [16]. In this research, the authors claimed that the types of the different maintenance tasks should be considered to improve the outcomes of the estimation model being used.…”

An Approach to Probabilistic Effort Estimation for Military Avionics Software Maintenance by Considering Structural Characteristics

“…where, y is an observed value of the dependent variable, y i is the value of the i − th value of the dependent variable as observed in the data set, y i * is the predictive value from the point forecast model trained with all the observations except the i − th, and n is a sample data set of size. Indeed, the literature and the experience show that accurate prediction is difficult: an average error of 100% can be considered "good" and an average error of 32% "outstanding" [12]. The measures resulting from cross-validation with statistics analysis are shown in Table 4.…”

Dynamics of software maintenance

“…It is also dependent on human interpretation and therefore prone to errors due to a host of technical, social and political reasons [9]. Software maintenance effort estimation is an important aspect of software maintenance planning [10]. Early and accurate estimates can significantly reduces risks and assist project managers in: • Improving maintenance process efficiency • Maintain vs. reengineer/reacquire decisions • Informed decision-making • Budgeting staff allocation and rotation Software maintenance broadly includes error corrections, changes (amendments or enhancements) and improvements to operational software.…”

“…One approach could be to statistically calculate the maintenance effort based on the effort expended in the other phases of the software life cycle; however, in reality this is usually not possible due to a lack (or absence) of such data for the development phases [9]. It is also important to appreciate that software development and maintenance are also different types of activities and have different inherent characteristics [10]. Software maintenance is an evolutionary process [20] and we believe that it is of equal if not of more fundamental importance to the software industry, and has probably not received the attention and respect that it deserves, especially in the context of estimation models.…”