Evolutional meta-learning framework for automatic classifier selection
Silviu Cacoveanu, Camelia Vidrighin, Rodica Potolea · 2009
Meta-learning is currently a hot research topic in machine learning, which has emerged from the need to support data mining automation in issues related to algorithm and parameter selection. Finding the best learning strategy for a new domain/problem can prove to be an expensive and time-consuming process even for the experienced analysts. This paper presents a new meta-learning system, designed to automatically discover the most reliable learning schemes for a particular dataset, based on the knowledge the system acquired about similar datasets. The novelty of the approach consists in combining dataset characterization with landmarking to increase the accuracy of the predictions. The proposed architecture is aiming to resolve the problem of selecting the best classifier for a dataset while minimizing the work done by the user but still offering flexibility.