Data-Centric Machine Learning: Enhancing Performance through Intelligent Data Curation

Abstract

In the evolving landscape of artificial intelligence, the emphasis on model-centric optimization has overshadowed the importance of data quality. Data-centric machine learning (DCML) challenges this narrative by shifting the focus from fine-tuning model architectures to improving the quality, consistency, and relevance of the data used to train these models. This research paper investigates the transformative impact of intelligent data curation on the performance of machine learning systems. We explore the principles and practices of DCML, evaluate various data curation techniques such as data augmentation, labeling optimization, deduplication, and error correction, and present experimental results showcasing performance improvements across multiple benchmark datasets. The findings underscore the critical role of high-quality data in achieving robust, reliable, and interpretable machine learning models.