Bio-inspired Algorithms in Software Fault Prediction: A Systematic Literature Review

Abstract

Bio-inspired (and meta-heuristic) algorithms are successfully employed in different domains and the research is going on to accommodate them in all the contexts where optimization is required. In software engineering, and especially Software Fault Prediction (SFP), they are investigated in various forms, e.g., to extract the most relevant features in a dataset or to select the most appropriate set of parameter values in the application of estimation techniques. In SFP, feature selection and optimization/tuning of estimation technique's parameters are an active research area, where recently various bio-inspired algorithms have been employed for both strategies. In this work, we present a Systematic Literature Review (SLR) about the use of bio-inspired algorithms for feature selection and parameter optimization aiming at increasing fault prediction accuracy of the models built with various estimation techniques. To the best of our knowledge, there is no SLR in SFP which covers the use of bioinspired algorithms, both for feature selection and parameter optimization. Since, the use of bio-inspired algorithms in the area of SFP started to be investigated in the late 2000, we have considered studies published between 2007 and 2019. As result, we have selected about 19 studies related to parameter optimization and 15 dealing with feature selection (in total 34 studies), extracted from five well-known digital libraries (ACM digital library, IEEE explore, Springer, ScienceDirect, and Scopus). Genetic Algorithms (GA) and Particle Swarm Optimization (PSO) are the widely used bio-inspired algorithms, both for parameter optimization and feature selection. Among them, GA is the better performed algorithm when evaluating its performance against the baseline (i.e., estimation techniques without any algorithm for feature selection or parameter optimization and trained with their default values). The SLR results also suggests that bio-inspired algorithms seem to provide more accurate predictions for feature selection than for parameter optimization.