Feature Subset Optimization through the Fireworks Algorithm

Software estimation models are vital in the software industry, given the fact that most software development projects over-run the time and budget limits. Recently, data mining algorithms have been applied to further improve the prediction accuracy of the software cost estimation models that are routinely used in industry. This paper introduces a novel Swarm Intelligence technique to fine-tune such models. It chooses the Feature Selection Method (FSS) to reduce the number of input parameters in the dataset. Further, it applies the Fireworks Algorithm (FWA) to deal with the combinatorial explosion problem in determining the optimal subset of features. The FWA-FSS method improves the prediction accuracy when tested on publicly available software development project datasets.