AI-Driven Software Quality Prediction: A Hybrid Deep Learning and Evolutionary Algorithm Approach

Abstract

Objectives:
This study aims to enhance software quality prediction by addressing the limitations of traditional machine learning models—namely, feature redundancy and high computational costs—through a hybrid deep learning approach integrated with Genetic Algorithms (GA) for efficient and accurate defect prediction.

Methods:
The proposed method combines Convolutional Neural Networks (CNNs) and Multilayer Perceptrons (MLPs) for deep feature learning, alongside GA for optimal feature selection. Model pruning and quantization techniques were employed to improve computational efficiency. The hybrid model was evaluated using publicly available software defect datasets.

Findings:
The hybrid approach achieved an accuracy of 89.2%, surpassing traditional classifiers like Random Forest and Support Vector Machines (SVMs). Furthermore, computational efficiency was improved by 35%, confirming the effectiveness of the model in balancing accuracy and performance.

Novelty:
Unlike conventional models, this approach integrates deep learning with evolutionary feature selection and computational optimization strategies, resulting in a robust, scalable, and efficient solution for software defect prediction. This combination ensures high predictive performance while minimizing resource consumption, making it suitable for large-scale and real-time applications.