An Evolutionary Approach for Multi-Objective Feature Selection in ADMET Prediction

SOTO, AXEL J.; CECCHINI, ROCÍO L.; PALOMBA, DAMIÁN; VAZQUEZ, GUSTAVO E.; PONZONI, IGNACIO

Santa Fé, Argentina

Conferencia; CLEI 2008 (XXXIV Conferencia Latinoamerica de Informática); 2008

CLEI (Centro Latinoamericano de Estudios en Informática)

This paper presents multi-objective evolutionary methods for determining the most relevant set of variables for predicting physicochemical properties. The multi-objective approach is useful to both minimize the cardinality of the subset as well as to maximize its predictive capacity. In this sense, rigorous experimentations were carried out in order to determine which multiobjective strategy is better for the feature selection task. Based on the results over a logP (octanol-water partition coefficient) data set, we may argue that the aggregative (non-Pareto) strategy constitutes a wise search strategy.