Comparing model-centric and data-centric approaches to determine the efficiency of data-centric AI
(1) VNU-HCM High School for the Gifted, Ho Chi Minh, Vietnam
https://doi.org/10.59720/22-112![Cover photo for Comparing model-centric and data-centric approaches to determine the efficiency of data-centric AI](https://emerginginvestigators.org/rails/active_storage/representations/proxy/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaHBBcHNOIiwiZXhwIjpudWxsLCJwdXIiOiJibG9iX2lkIn19--6cb52032efa80660f423b2dd3af034f3e8d9080f/eyJfcmFpbHMiOnsibWVzc2FnZSI6IkJBaDdCem9MWm05eWJXRjBTU0lJY0c1bkJqb0dSVlE2QzNKbGMybDZaVWtpRFRZd01IZzJNREErQmpzR1ZBPT0iLCJleHAiOm51bGwsInB1ciI6InZhcmlhdGlvbiJ9fQ==--33b2b080106a274a4ca568f8742d366d42f20c14/figures.png)
In current machine learning approaches, data is crucial, yet it is often overlooked and mishandled in artificial intelligence (AI). As a result, many hours are wasted fine-tuning a model based on faulty data. Hence, there exists a new trend in AI, which is data-centric AI. We hypothesized that data-centric AI would improve the performance of a machine learning model. To test this hypothesis, three models (two model-centric approaches and one data-centric approach) were used. The model-centric approaches included basic data cleaning techniques and focused on the model, while the data-centric approach featured advanced data preparation techniques and basic model-training. We found that the data-centric approach gave a higher accuracy than the model-centric approaches. The model-centric approaches achieved 91% and 90% accuracy, respectively, whereas the data-centric approach achieved 97% accuracy.
This article has been tagged with: