Elena Pokryshevskaya, Evgeny Antipov
In this study, nine methods for measuring indirect importance are compared on the basis of their discriminatory power and stability. To the best knowledge of the authors, the stability of results obtained with different methods is assessed for the first time. The deficiencies of an existing criterion for assessing diagnosticity are pointed out and a modified version suggested. The empirical comparison is based on two real-world datasets from the ecommerce industry. Even though none of the methods appeared to be the best according to both criteria simultaneously, there seem to be grounds for recommending the theoretically sound Shapley value decomposition of T-square if stability and discrimination are about equally important for a decision maker, while negative contributions are undesirable
© 2001-2024 Fundación Dialnet · Todos los derechos reservados