Research on Intelligent Decision-Making AI Algorithms for Software Defect Identification

Abstract

With the continuous growth in the scale and complexity of software systems, traditional defect identification methods are becoming insufficient to meet the needs of modern software development. Intelligent decision-making algorithms based on machine learning and deep learning have demonstrated significant advantages in the field of software defect identification. Through comparative analysis of multiple algorithms—such as deep neural networks, support vector machines, and random forests—in defect identification, the study shows that deep learning models incorporating attention mechanisms can increase defect identification accuracy to 92.8%, which is 15.6 percentage points higher than traditional methods. Experimental results confirm that an integrated learning framework, combined with code feature extraction and defect pattern analysis, can effectively enhance software quality assurance efficiency. By employing techniques such as abstract syntax trees and program dependency graphs, automatic extraction of code features was achieved. Furthermore, using a multi-head attention mechanism enhanced the feature representation capability, ultimately constructing an end-to-end intelligent defect identification system.