Objective Ridge Regression System

In the actual system, it is very important to get enough training samples for the performance of the model. However, no matter how rich the collection is, there will be a problem: limited training sets cannot really explain the label of these. So we propose an Objective Ridge Regression System (ORRS) to solve this problem in the classification stage. First of all, according to different categories, training samples are regressing to the different objectives we defined by, to obtain a transform matrix. Secondly, all of the test samples are projected into the low dimensional space by this transform matrix. Thirdly, we developed a Diversity Factors Rule (DFR) to divide test samples into multiple batches. Finally, the Minimum Distance Classifier algorithm is used to classify each batch of test samples, and the class-centers are updated synchronously. Circulate the previous steps until all batch samples are classified. Experimental results on the Iris, Wine, ORL and AR databases show that our method outperforms some other algorithms.