FEATURES OF APPLYING NEURAL NETWORKS FOR PREDICTING SOFTWARE QUALITY METRICS

Abstract

This article explores the application of neural networks for predicting software quality metrics. The primary focus

is on the SHAP (SHapley Additive exPlanations) method, one of the most advanced tools for interpreting machine learning models. Other interpretation methods such as LIME (Local Interpretable Model-agnostic Explanations), Integrated Gradients, DeepLIFT (Deep Learning Important FeaTures), Grad-CAM (Gradient-weighted Class Activation Mapping), Feature Ablation, and Permutation Feature Importance are also considered. Significant contributions to the development of theoretical and practical aspects of generalizing the issues related to the application of neural networks for predicting software quality metrics have been made by scientists such as Poligné, I., Broyart, B., Trystram, G., Rahman, N. H. A., Lee, M. H., and others. The aim of the article is to investigate the theoretical and methodological features of applying neural networks to predict software quality metrics. The article presents a detailed analysis of each method, including their advantages and disadvantages. To achieve this aim, the methods were compared based on their ability to ensure model transparency and prediction accuracy; practical aspects of implementing neural networks in the process of predicting software quality were discussed, along with the advantages and disadvantages of each method in the context of real-world tasks; a practical solution for improving the SHAP method was developed. The results obtained underscore the importance of choosing the appropriate interpretation method to ensure reliable and understandable software quality predictions. The process of addressing the raised issues involved the use of analysis, synthesis, generalization, and comparison methods.

Keywords: SHapley Additive exPlanations, Deep Neural Networks, Convolutional Neural Networks, Recurrent Neural Networks, Gradient-weighted Class Activation Mapping, software systems quality assessment.