A Self-Learning Framework for Enhancing Data Quality in Data-Centric AI Systems through Continuous Feedback

Data quality is of prime importance for the feedback AI system in offering consistent and stable performance. Practical application datasets stand to be most affected with noise, missing values, inconsistency, and data drift and consequently tend to have negative impacts on model accuracy as well as generalizability. This paper has proposed a new paradigm of self-upgrading data quality of feedback AI systems forever through reinforcement learning-based optimization, adaptive preprocessing, and self-strengthening quality estimation. It has an end-to-end measurement scheme for data quality, real-time feedback loop model, and dynamic data preprocessing strategy with adaptive self-learning mechanism with real-time estimates. Preprocessing operations are scaled up through Q-learning so that the system can learn and scale. Large-scale evaluations on benchmark datasets such as CIFAR-10 and OpenML achieve 15-20% model accuracy improvement and 30% reduction in data preprocessing error compared to state-of-the-art techniques such as CleanLab and Snorkel. Simulation with real-world usage cases in finance and medicine offers the effectiveness of the framework with 22-25% improvement in data quality metrics. The proposed framework is an automated, scalable solution to a timely data quality issue in data-intensive AI and holds promise for use in dynamic high-stakes scenarios. Multi-modal fusion and cross-domain scalability are the focus of future efforts.