Dynamic Class Prediction with Classifier Based Distance Measure

Combining multiple classifiers (ensemble of classifiers) to make predictions for new instances has shown to outperform a single classifier. As opposed to using the same ensemble for all data instances, recent studies have focused on dynamic ensembles in which a new ensemble is chosen from a pool of classifiers specifically for every new data instance. We propose a system for dynamic class prediction based on a new distance measure to evaluate the distance among data instances. We first map data instances into a space defined by the class probability estimates from a pool of two-class classifiers. We dynamically pick classifiers (features) to be used and the k-nearest neighbors of a new instance by minimizing the distance between the neighbors and that instance in a two-step framework. Results of our experiments show that our measure is effective for finding similar instances and our framework helps making more accurate predictions.