A Hybrid Neuro-Fuzzy Machine Learning Framework for Predicting Software Quality in Open-Source Projects

Abstract: The increasing complexity of software systems, especially in Free and Open-Source Software (FOSS) ecosystems, there has been a growing requirement for the adoption of automated software quality prediction tools. In this context, this study attempts to present a hybrid Neuro-Fuzzy Machine Learning (NF-ML) model to predict software quality by identifying defective software modules. In this study, the proposed model combines the efficient learning capability offered by neural networks and the malleable reasoning capabilities offered by fuzzy logic to create a robust model with high interpretability. In this experiment, the NASA Metrics Data Program (MDP) dataset sourced from Kaggle was used to train and test the proposed model. Additionally, data pre-processing was conducted to address missing data points and normalize the data points using chi-squared feature selection. The proposed model was developed using TensorFlow programming from Python with various performance metrics such as accuracy, precision, recall, F1-measure, and AUC-ROC accuracy to determine efficiency. The experimental outcome of this research study suggests that the hybrid NF-ML model attains better efficiency than ML models to predict software quality while increasing reliability.