Rot-SiLA: A Novel Ensemble Classification Approach Based on Rotation Forest and Similarity Learning Using Nearest Neighbor Algorithm

Recent years have seen a great inclination towards Machine Learning classification and researchers are thinking in terms of achieving accuracy and correctness. Many studied have proved that an ensemble of classifiers outperform individual ones in terms of accuracy. Qamar et al. have developed a Similarity Learning Algorithm (SiLA) based on a combination of k nearest neighbor algorithm and Voted Perceptron. This approach is different from other state of the art algorithms in the sense that it learns appropriate similarity metrics rather than distance-based ones for all types of datasets i.e. textual as well as non-textual. In this paper, we present a novel ensemble classifier Rot-SiLA which is developed by combining Rotation Forest algorithm and SiLA. The Rot-SiLA ensemble classifier is built upon two types of approaches, one based on standard kNN and another based on symmetric kNN (SkNN), just as was the case with SiLA algorithm. It has been observed that Rot-SiLA ensemble outperforms other variants of the Rotation Forest ensemble as well as SiLA significantly when experiments were conducted with 14 UCI repository data sets. The significance of the results was determined by s-test.