Ensemble learning via feature selection and multiple transformed subsets: Application to image classification

In the machine learning field, especially in classification tasks, model’s design and construction are very important. Constructing model via a limited set of features may sometimes bound performance lead to non-optimal performances that some algorithms can provide. To this end, Ensemble methods were proposed literature. These methods’ main goal is learn models provide or predictions whose joint use could better than obtained by single model. paper, we propose three variants new efficient ensemble approach was able enhance linear discriminant embedding method. As case study consider “Inter-class sparsity discriminative least square regression” We seek estimation an enhanced data representation. Instead deploying multiple classifiers on top transformed features, target extracted feature subsets learned embeddings. associated with ranked original features. Multiple used for estimating transformations. The derived concatenated form representation vector process. Many factors studied investigated paper including (Parameter combinations, number models, different training percentages, selection etc.). Our has been benchmarked image datasets various sizes types (faces, objects scenes). scheme achieved competitive four face (Extended Yale B, LFW-a, Gorgia FEI) as well COIL20 object dataset Outdoor Scene dataset. measured our schemes comparison (the ICS_DLSR, RDA_GD, RSLDA, PCE, LDE, LDA, SVM KNN algorithm) conducted experiments showed manner compared single-model based outperform its competing methods.