Integration of Machine Learning and Large Language Models for Scalable Data Cleaning

Data quality is the measure to which data satisfy the demands of its use in accuracy, completeness, consistency, timelines, and reliability. Data quality problems such as missing values, outliers, and semantic inconsistencies—within structured databases seriously hinder good data analysis and decision-making in mission-critical applications. This paper introduces a hybrid AI-driven framework integrating classical machine learning with current large language models for autonomous and also scalable data cleaning. Specifically, Random Forest and K-Nearest Neighbours are employed for precise missing or outlier value identification and imputation, and pretrained transformer-based models such as BERT and GPT-3.5 are employed for outlier and semantic discrepancy identification and correction in tabular data. The proposed framework is evaluated using public and real-world datasets in the healthcare, finance, and retail sectors. Experimental findings demonstrate an optimal 20% improvement in overall data quality over conventional rule-based and statistical cleaning techniques. The results highlight the versatility, precision, as well as domain-agnostic characteristics of the proposed solution, placing AI technologies as essential building blocks in next-generation data maintenance pipelines for modern data-driven applications.