An Improved Genetic Algorithm for Multi-Objective Test Suite Optimization in Agile Software Development

Regression testing in agile development requires selecting a compact, fault-effective subset of test cases from a growing suite within strict sprint time constraints. This challenge is a multi-objective combinatorial optimization problem requiring simultaneous maximization of fault detection rate (FDR), minimization of execution cost, and maintenance of code coverage breadth. Standard single-objective approaches sacrifice one dimension, while NSGA-II ignores structural dependency relationships among test modules.This paper proposes IGA-MO (Improved Genetic Algorithm for Multi-Objective optimization), extending NSGA-II with three innovations: (1) a semantic-aware crossover that builds a dependency graph from Jaccard similarity of McCabe-Halstead metric vectors and exchanges connected component blocks; (2) an adaptive mutation mechanism driven by Hamming distance diversity monitoring; and (3) a bounded elite archive with crowding-distance truncation. A complete Algorithm 1 pseudocode and full equation explanations are provided.Experiments use the NASA CM1 dataset from the PROMISE Repository — 498 real C-language software modules from NASA satellite flight software with 49 NASA-verified defects (9.8%) and 21 McCabe and Halstead metrics. Over 10 independent runs, IGA-MO achieves a mean FDR of 63.1% (±18.4%) with 60.5% suite reduction, significantly outperforming random selection (Wilcoxon W=50.0, p=0.0098). All results are fully measured on real NASA data.