On the application of search-based techniques for software engineering predictive modeling: A systematic review and future directions

Threats to validity in search‐based predictive modelling for software engineering

“…What are the various SBAs applied to software predictive modelling? : According to Malhotra et al [1], SBAs which are used for developing SPMs can be categorised into four broad categories: local search, evolutionary, swarm intelligence and hybrid. A local search technique starts its search from a specific candidate solution and explores only its neighbourhood to search for an optimal solution.…”

“…• Many SBAs such as GA, genetic programming and particle swarm optimisation do not require any pre-assumptions about the training data. This characteristic makes them suitable candidates for developing prediction models from historical training data, as the data does not need to follow specific assumptions (S10, S12, S15, S21, S48 Apart from these general characteristics, other specific characteristics of a particular SBA may aid its applicability for developing SPMs in a certain scenario [1].…”

“…The results of the review are based on the assumption that the selected primary studies are unbiased. Furthermore, our review does not include any unpublished work [1], which could pose a possible threat to the obtained conclusions. As the study reports threats which were mentioned in two or more studies, the obtained results are unbiased and are not specific to a study's experimental design.…”

“…A recent review by Malhotra et al [1] was performed on 78 primary studies, which use SBAs for effort estimation, defect proneness prediction, maintainability prediction and change prediction. However, the review did not evaluate the concerns to validity in these studies.…”

“…SPMs can be used for resource planning, and optimisation of schedule and cost of software projects as they enable forecasting of various software attributes at the early stages of the software development lifecycle. Thus, corrective actions can be taken in order to produce effective, good quality and maintainable products [1]. In order to develop SPMs, a classification/regression technique is used which uses historical data from the software projects for training, and thereby forecasts the specific software attribute of upcoming releases of the software product.…”