Akademik Veri Yönetim Sistemi
SOFTWARE QUALITY JOURNAL, cilt.34, sa.2, 2026 (SCI-Expanded, Scopus)
Software testing processes heavily rely on the quality and realism of test data. Traditional approaches often face limitations due to static or insufficient data, failing to capture complex real-world user behaviors, especially in dynamic domains like mobile advertising. To address this challenge, this paper presents a novel AI-based framework designed to autonomously generate and assure the quality of synthetic data intended for use in testing processes. The system integrates a GAN-inspired feedback-driven generative framework based on MLP regression models for generating realistic tabular data mimicking mobile advertising logs. The self testing mechanism of the proposed AI-based framework evaluates the functional realism of synthetic data using Random Forest and XGBoost machine learning models to compare predictive performance. This performance is quantitatively measured by the Mean Squared Error (MSE) metric. If the mechanism detects inadequate data quality (high MSE), an automated feedback loop activates GAN retraining, enabling continuous improvement. Experimental results demonstrate the effectiveness of the proposed system in generating high-quality synthetic data that closely aligns with real-world data characteristics. This study contributes a fully automated, feedback-driven synthetic data generation and quality assurance system, enhancing the potential for more robust and realistic software testing by providing validated test inputs applicable to domains with dynamic user behavior.