Research On The Impact Of The Improved Sample Selection On Classification Algorithm

Machine learning is a mechanism on the machine to understand the behavior and machine learning capabilities have, subject to build self-learning function of a computer program. In recent years, machine learning theory has been successfully applied in many applications and development, has become one of the hot base and computer science. Machine learning has been a very wide range of applications such as search engines, marketing, network traffic monitoring, stock market, and voice recognition. Those who have experienced the presence of local, machine learning methods can play a very important role in it.Supervised learning is the use of a set of samples of known classes of adjustment classifier parameters to achieve the desired properties of the process. Supervised learning has major problems that need attention bias variance trade-off, functional complexity and number of training data, dimension, such as noise in the output value of the input space. This paper mainly include verification of supervised learning neural networks, decision trees, etc.Feature selection and sample selection is mainly used to solve the dimensionality of samples in the learning process is too high or too large sample size problem. Feature selection can eliminate not want to control or hyperactivity than traits, thus to reduce the number of features to improve model accuracy, reducing the running time of entry. And if we can reduce the sample selection based on the study sample did not affect the original index, it will definitely be useful, in addition, the sample selection if they can recommend a better sample for classification, must also be able to get very excellent classifier.This paper firstly introduces the related knowledge of sample selection, sample selection leads to the main working process, methods for the different classification method and difference, for different categories of analysis, leads to the sample selection method. Because the sample selection based on classifier before, so this paper generally introduces the classification model of the mainstream, then with corollary two main model of this paper need to experiment, respectively, neural network and decision tree, and to explore the method of sample selection and classifier with what kind of combination. The sample selection method to design the heuristic algorithm, first demonstrates the correctness of the theory, and finally the simulated annealing approach to heuristic algorithm as the core.We put forward a selection method based on improved heuristic algorithm is used to sample, an improved neural network and decision tree algorithm, after a lot of experiments on the improved algorithm, carries on the contrast experiment, summed up the advantages and disadvantages of the algorithm in the improved algorithm, finally discusses the prospects and practical application.