Evolutionary ensembles based on prioritized aggregation operator

Ensemble methods are advanced learning algorithm proposed for generating base classifiers and accumulating them all together to derive a new classifier which is expected to perform better than the constituent classifier. This study proposes a novel ensemble technique where a base learning classifier is trained repeatedly by using different weightings over the training samples or examples, and the process is governed by the conceptualization of evolutionary processes and the aggregation operators. We utilize the evolutionary technique that can efficiently search a large weighing space for enriching suitable weights (chromosome) to the training samples. For finding an appropriate weighting, the crossover and mutation processes are applied on the weighting space to get the optimized set of weights which is accomplished through different generations. The considered base learning classifier is trained over the training examples along with their respective weightings by utilizing a learning algorithm, and for the finite number of generations, the weights are evolved and optimized through the evolutionary process. All the classifiers obtained in different generations of the evolutionary process are utilized for efficiently building the final ensemble. The set of classifiers obtained in different generations are combined together by utilizing the concept of priority-based averaging aggregation operator by availing priority to different generations. The classifier ensemble is done with two forms of operators: one without priority degree and the other with the priority degree. The proposed classifier ensemble algorithm is tested over the UCI benchmark dataset. The results obtained through the experimental process are more accurate, consistent, and reliable while comparing to other state-of-the-art methods, which ensures the efficacy of the proposed algorithm.