Linear Least-Squares Fusion of Multilayer Perceptrons for Protein Localization Sites Prediction

摘要

This paper presents a new type of linear model of fusing multilayer perceptrons for predicting protein localization sites. The Linear Least-Squares Fusion (LLSF) model makes a set of component networks work collectively and integrates their knowledge in order to ameliorate the generalization capability of a classification system. The empirical results show that the LLSF system reached an overall accuracy of 85.4% in predicting 336 E.coli proteins, better than the performance of its component networks or the previous method in literature.