Data valuation using Shapley value in machine learning

• Main Authors: Sharma, Pushti, Singh, Gurpreet, Kaur, Navneet
• Format: Conference Proceeding
• Language: English
• Subjects: Game theory; Machine learning
• ISSN: 0094-243X; 1551-7616
• DOI: 10.1063/5.0178097

To measure the value of the information in effective way and choices would be a key problem as data would become the fuel supporting both technological and economic advancement. Many methodologies have already been offered further to attribute a machine learning model’s prediction to those same required input characteristics in order to demonstrate it. Technologies that use the Shapley value approach from cooperating game theory are common among them. While previous studies have concentrated on the theoretical basis for Shapley values as well as effective methods for computing them, they have provided insufficient support for both the game models employed. Shapley value has several uses in machine learning. In this paper conceptual framework for creating and analyzing explanation of ML models trained on large datasets, have really been extracted from the well-known Kaggle site and that data Shapley has several other benefits in term of computational contribution for effective data valuation, local interpretability along with Global model interpretability.