Meta-inductive prediction based on Attractivity Weighting: Mathematical and empirical performance evaluation

Publication date: April 2019Source: Journal of Mathematical Psychology, Volume 89Author(s): Paul D. Thorn, Gerhard SchurzAbstractIn the present paper, we present mathematical and empirical results concerning the performance of a meta-inductive prediction method known as Attractivity Weighting. The mathematical results show that Attractivity Weighting is endowed with important guarantees concerning its worst-case short run and long run performance. In addition to these guarantees which hold for all logically possible environments, simulations applied to data describing real-world environments suggest that the short run performance of carefully selected forms of Attractivity Weighting is generally very good, both in environments in which one-reason prediction methods are optimal and in environments in which weighting methods are optimal. In both sorts of environment, Attractivity Weighting approximates the performance of that available prediction method that is optimal in that environment.
Source: Journal of Mathematical Psychology - Category: Psychiatry & Psychology Source Type: research