Paper Contents
Abstract
Software bugs pose significant challenges in software engineering, leading to security issues, system failures, and high operational costs. Traditional bug detection methods like static and dynamic analysis suffer from limitations such as high false positives, slow processes, and reliance on human expertiseissues that worsen as software complexity grows.Deep learning has emerged as a powerful solution for automated bug detection and resolution. Unlike rule-based methods, deep learning models can learn from historical code to identify patterns and anomalies without predefined rules. These models analyze source code, logs, and system behavior to detect bugs with greater accuracy and efficiency.Neural network architectures such as CNNs (for token-based analysis), RNNs and LSTMs (for sequential dependencies), and transformer-based models like CodeBERT and GPT-4 have shown exceptional performance in this domain. Transfer learning and reinforcement learning further enhance model adaptability and decision-making in real-world scenarios.Despite progress, challenges remain, including the need for large, labeled datasets, model interpretability, and high computational costs.This paper offers a comparative analysis of deep learning models for bug detection, evaluating them based on accuracy, false positives, and efficiency. It also introduces a hybrid approach that combines multiple deep learning techniques to maximize performance while addressing individual model limitations.In conclusion, deep learning is transforming software debugging, reducing defects, and improving development efficiency, with growing potential for seamless integration into modern development pipelines.
Copyright
Copyright © 2025 Anunay Anand. This is an open access article distributed under the Creative Commons Attribution License.
