An Exploratory Study for Predicting Maintenance Effort using Hybridized Techniques
Ruchika Malhotra, Kusum Lata · 2017
Software maintenance effort prediction is one of the very costly and challenging affair in the process of software development. Early detection of changes in software are also necessary as it helps the software developers and project managers to allocate resources in an efficient manner. It is very critical for the project managers and software developers to detect changes in software in the earlier phases of software development so that the portions of software that are more prone to changes can be restructured and redesigned. Various statistical and machine learning based models are available for maintenance effort prediction of these objects oriented systems. In this paper, we propose a novel approach for maintenance effort prediction using hybridized (i.e., combining search-based techniques with machine learning alternatives) techniques. Specifically, we will address these research issues: (i) low repeatability of empirical studies involving maintainability models, (ii) less usage of statistical tests for comparing the effectiveness of different models, and (iii) non-assessment of predictive performance of hybridized techniques. The models are constructed using object-oriented metrics and the results of this research are validated using two commercial datasets. Based on the experiments conducted it has been proved that hybridized techniques have capability for predicting maintenance effort.