Machine Learning Techniques for Cleaning Raw Data

Raw data collected from real-world environments is often incomplete, noisy, and inconsistent, significantly affecting machine learning (ML) model performance. This paper proposes a hybrid data cleaning framework that integrates statistical preprocessing with machine learning-based anomaly detection techniques. Unlike traditional approaches, the proposed method combines Z-score normalization, K-Means clustering, and Isolation Forest for robust outlier detection and data refinement. Experimental evaluation is conducted on larger benchmark datasets, and results are compared with baseline statistical methods. The proposed approach improves model accuracy, data consistency, and overall data quality. The findings demonstrate that hybrid intelligent data cleaning significantly enhances downstream ML performance and scalability.