Research Of Classification Algorithm Based On K Nearest Neighbor

Data mining is a widely field of machine learning, and it integrates the artificial intelligence technology and database technology. It helps people extract valuable knowledge from a large data intelligently and automatically to meet different people applications. KNN is a used method in data mining based on Statistic. The algorithm has become one of the ways in data mining theory and application because of intuitive, without priori statistical knowledge, and no study features .The main works of this thesis is k nearest neighbor classification algorithm. First , it introduces mainly classification algorithms of data mining and descripts theoretical base and application. This paper points out the reasons of slow and low accuracy and proposes two improved ways.In order to overcome the disadvantages of traditional KNN, This paper use two algorithms of classification and clustering to propose an improved KNN classification algorithm . Experiments show that this algorithm can speed up when it has a few effects in accuracy.According to the problem of classification accuracy, the paper proposes a new calculation of weight. KNN the traditional method generally used Continental distance formula measure the distance between the two samples. As the actual sample data collection in every attribute of a sample of the contribution is not the same, often using the weighted Continental distance formula. This paper presents a calculation of weight , that is weighted based on the characteristics of KNN algorithm. According to this Experiments on artificial datasets show that this algorithm can improve the accuracy of classification.Last, the paper indicates the direction of research in future based on the full-text.