ANALYZING MULTI OBJECTIVE SEARCH ALGORITHMS TO PREDICT FAULTY CLASSES

Abstract

Many real world problems involve optimization of multiple objective functions on a feasible variable space. These objective functions are often conflicting and cannot be formulated as a scalar function. Such problems are known as multi-objective optimization (MOO) problems as there are multiple objectives which need to be optimized simultaneously. A recent MOO problem in software engineering domain is the prediction of faulty classes in a software. While faulty classes are predicted, finding a trade-off between two conflicting objectives is essential. The first one is minimizing the number of classes to be recommended and the second one is maximizing the relevance of the solution which is based on the history based and lexical based similarities between the Application program interface (API) document and the bug description. Evolutionary algorithms (EA) seem to be well suited to solve such MOO problems as they parallely generates a set of solutions, ultimately exploiting similarities of solutions by crossover. Previous studies have suggested that EAs show improved performance over other search algorithms for solving MOO problems. This study evaluates the use of two EA’s namely the Non-dominated Sorting Genetic Algorithm (NSGA-II) and the Strength Pareto Evolutionary Algorithm 2 (SPEA2) to predict faulty classes. The results are empirically validated on six open source Java projects. They point towards the superiority of the NSGA II algorithm over the SPEA 2 algorithm.