Three empirical studies on predicting software maintainability using ensemble methods

Abstract

More accurate prediction of software maintenance effort contributes to better management and control of software maintenance. Several research studies have recently investigated the use of computational intelligence models for software maintainability prediction. The performance of these models, however, may vary from dataset to dataset. Consequently, ensemble methods have become increasingly popular as they take advantage of the capabilities of their constituent computational intelligence models toward a dataset to come up with more accurate or at least competitive prediction accuracy compared to individual models. This paper investigates and empirically evaluates different homogenous and heterogeneous ensemble methods in predicting software maintenance effort and change proneness. Three major empirical studies were designed and conducted taken into consideration different design such as the types of the investigated ensembles methods, types of prediction problems, used datasets, and other experimental setup. Overall empirical evidence obtained from the three studies confirms that some ensemble methods provide more accurate or at least competitive prediction accuracy compared to individual models across datasets, and thus they are more reliable.

Self-learning Change-prone Class Prediction

“…They found that size metrics were more discriminative that other OO metrics, such as cohesion, coupling and inheritance. Elish et al [5] proposed an empirical study which used ensemble methods on change prediction. They found that ensemble methods can achieve a better performance than individual models.…”

“…The percentage and the number of instances possess a substantial range which can validate the model ability among a wide range. Considering the code metrics, we adopt the typical metrics which are identical with the relevant studies of change prediction [1,5,11,32] as the Table II shows. In detail, five Chidambar and Kemerer metrics [33]: WMC, DIT, NOC, RFC, and LCOM; four Li and Henry metrics [34]: MPC, DAC, NOM, SIZE2; and one traditional lines of code metric (SIZE1) are adopted.…”

“…The first baseline is the change-prone class prediction based on logitboost (LB) which is proposed by Malhotra et al [13]. In addition, to avoid the bias from only one method, we also adopt three typical machine learners as baselines which are used in all three empirical studies on change prediction proposed by Elish et al [5]. Therefore, the second, third and the fourth baseline is Multilayer perceptron (MLP), Radial basis function network (RBF) and Support vector machine (SVM), respectively.…”

“…It can represent the weak part of a software system [2]. Thus, software change-prone class prediction contributes to better allocation of software resources (e.g., time and staff) in the software maintenance process [5]. This technique aids to support maintenance related decision making by identifying change-prone classes in advance.…”

“…[6], code smells [7], design patterns and [8] evolution metrics [9,10]. In terms of the techniques, different machine learning approaches have been used, such as Bayesian networks [11], neural networks [12], multivariate regression [1] and ensemble methods [5]. A typical prediction model based on machine learning is designed by learning from a historical labeled dataset in a supervised way.…”