A resilient voting scheme for improving secondary structure prediction

This paper presents a novel approach called Resilient Voting Scheme (RVS), which combines different predictors (experts) with the goal of improving the overall accuracy. As combining multiple experts involves uncertainty and imprecise information, the proposed approach cancels out the impact of bad performers while computing a single collective prediction. RVS uses a genetic algorithm to assign a reliability to each expert, by using the Q3 measure as fitness function. A resilient voting is then used to improve the accuracy of the final prediction. RVS has been tested with well known datasets and has been compared with other state-of-the-art combination techniques (i.e., averaging and stacking). Experimental results demonstrate the validity of the approach.