Review of AI-Driven Approaches for Automated Defect Detection and Classification in Software Testing

Increased size and complexity of modern software systems have necessitated novel, smart approaches to detecting and classifying defects. Within this review, we provide a comprehensive perspective on artificial intelligence (AI)-driven techniques within the umbrella of automated software testing. Relying on up-to-date research in machine learning, deep learning, and natural language processing, we explore how these technologies enhance accuracy, efficiency, and scalability of defect identification approaches. Studies such as VulDeePecker and other deep learning frameworks have revealed significant vulnerability detection improvements through automated static and dynamic code analysis. Systematic reviews and empirical evaluation of AI-based test automation tools also reveal their effectiveness in reducing human effort and improving software reliability. We structure these approaches by techniques (e.g., convolution neural networks, recurrent neural networks, transformers), application domains (e.g., source code analysis, PCB defect detection, test case generation), and performance metrics. The findings suggest a trend in the direction of integrated AI models for continuous testing, early-defect detection, and adaptive test generation. While there is an encouraging improvement, concerns regarding data availability, interpretability of models, and integration with current systems persist. This work concludes by proposing future research directions that include applying explainable AI and transfer learning to further improve automated software testing capabilities. Keywords: Artificial Intelligence (AI), Software Testing, Defect Detection, Defect Classification, Test Automation, Machine Learning, Bug predic-tion