A systematic literature review on empirical studies towards prediction of software maintainability

Abstract

Software maintainability prediction in the earlier stages of software development involves the construction of models for the accurate estimation of maintenance effort. This guides the software practitioners to manage the resources optimally. This study aims at systematically reviewing the prediction models from January 1990 to October 2019 for predicting software maintainability. We analyze the effectiveness of these models according to various aspects. To meet the goal of the research, we have identified 36 research papers. On investigating these papers, we found that various machine learning (ML), statistical (ST), and hybridized (HB) techniques have been applied to develop prediction models to predict software maintainability. The significant finding of this review is that the overall performance of ML-based models is better than that of ST models. The use of HB techniques for prediction of software maintainability is limited. The results of this review revealed that software maintainability prediction (SMP) models developed using ML techniques outperformed models developed using ST techniques. Also, the prediction performance of few models developed using HB techniques is encouraging, yet no conclusive results about the performance of HB techniques could be reported because different HB techniques are applied in a few studies.