FPAFS: Feature Selection Using the Flower Pollination Algorithm for Software Fault Detection System

This paper suggests a suitable feature selection (FS) approach FSFPA using flower pollination algorithm (FPA). It is based on the concept of flower pollination, to choose a set of important features or variables from a vast set of characteristics present in a dataset. The objective is to choose the best suitable features for defect prediction. In this study FSFPA strategy is compared with other optimization methods such as ant colony optimization (ACO), particle swarm optimization (PSO) and evolution based algorithms differential evolution (DE), and genetic algorithm (GA). The experiment has been conducted using these five optimization strategies on nine datasets. To determine the most effective metrics for identifying error proneness, it uses five classifiers (NB, KNN, DT, LDA, QDA). It is observed that FSFPA was better considering the parameters training times and prediction accuracies as compared to the traditional GA, ACO, DE, and PSO methods. In conclusion, FPA produces better prediction outcomes in imbalanced data sets than the GA, ACO, DE, and PSO. It is necessary to correct the parameters that are responsible for software errors by utilizing optimization algorithms and appropriate classifiers.