Turk, Salih; Coskuncay, Ahmet

Context: Automated detection and recommendation of software design patterns is crucial for software maintenance, reverse engineering, and architectural quality. Despite decades of research, the field has undergone a significant transformation with the emergence of deep learning and large language models (LLMs), leading to a fragmented landscape of methodologies and evaluation settings. Objective: This study aims to provide a systematic and evolutionary synthesis of design pattern detection and recommendation approaches published between 2015 and 2025, categorizing them into distinct technological paradigms and identifying open research challenges. Method: Following Kitchenham’s evidence-based software engineering guidelines and PRISMA principles, a systematic literature review was conducted. We executed a structured search on the Web of Science database, identifying 212 initial records. After rigorous multi-stage screening and quality assessment, 66 primary studies were selected for final synthesis and analysis. Results: The evolution of the field is characterized through six distinct paradigms: Rule-Based, Graph-Based, Feature-Driven Machine Learning, Representation Learning (Deep Learning), Hybrid, and Contextual Intelligence (LLM). Findings reveal a clear methodological shift from structural exactness to semantic and contextual understanding. However, the analysis also identifies a high concentration on specific datasets (JHotDraw and P-MARt) and a lack of standardized benchmarks for variant detection and multi-modal analysis. Conclusions: While LLM-based approaches show promise in handling pattern variants and semantic context, the field requires more diverse benchmarks and explainable models. This review provides a comprehensive research roadmap, emphasizing the need for multi-modal frameworks and industry-scale validation to bridge the gap between academic research and practical software engineering.

Turk, Salih; Coskuncay, Ahmet

Software defect prediction is essential for enhancing software quality by detecting software errors early in the development lifecycle. This article presents the usage and a detailed evaluation of Kolmogorov-Arnold Networks (KAN) and Multilayer Perceptrons (MLP) over eleven prevalent software defect prediction datasets, namely CM1, JM1, KC1, KC3, MC1, MC2, MW1, PC1, PC3, PC4, and PC5. The defect prediction models were established and evaluated using five principal performance metrics: Accuracy, Precision, Recall, F1 Score, and Area Under the Curve (AUC). The experimental findings indicate that KAN consistently surpasses MLP in all datasets and metrics. KAN attains an average accuracy enhancement of almost 12% compared to MLP and demonstrates substantial improvements in precision and recall, rendering it more dependable in detecting defected modules. The paired t-test findings indicate that the performance disparities between KAN and MLP are statistically significant (p-value< 0.05) for the majority of metrics. Also SHAP analysis was performed to determine the important rankings of features for all datasets according to the FastKAN model. This study underscores the promise of KAN as a superior alternative to traditional neural networks for software defect prediction. The results highlight the significance of the usage of sophisticated neural network topologies in tackling intricate software engineering challenges.

Turk, Salih; Bingol, Ozkan; Coskuncay, Ahmet; Aydin, Tolga

X-ray imaging is widely utilized for the detection of bone fractures due to its affordability, rapid processing capabilities, broad accessibility, and ease of use. Despite these advantages, the intricate analysis of X-ray images necessitates advanced computational techniques to fully exploit their rich informational content. Notably, accurate segmentation of these images plays a critical role in aiding medical professionals with precise diagnoses and effective treatment planning. This study examines the impact of integrating different backbone architectures for the task of fracture segmentation in X-ray images. Specifically, the research focuses on enhancing the widely-used DeepLabV3 model by incorporating pre-trained networks such as ResNet50, ResNet101, and MobileNetV3 into the encoder component to improve feature extraction and segmentation accuracy. The FracAtlas dataset, which presents unique challenges due to its small size and the diversity of fractures from various anatomical regions, was employed for model evaluation. Data augmentation techniques were implemented to expand the dataset, and an additional subset focusing on cropped images of fracture areas was developed. The models were trained over 50 epochs, and their performance was assessed using metrics such as Intersection over Union (IoU), loss values, and Dice scores. The results indicate that the DeepLabV3 model with ResNet-based backbones achieved IoU values exceeding 0.93 on the original dataset and demonstrated outstanding performance on the augmented and cropped datasets, with AUC values reaching up to 0.99. The study also highlights the computational complexity of the models, with ResNet101 exhibiting the highest time complexity, while MobileNetV3 was the most efficient in terms of processing time and memory consumption.