A Suspected Noise Filtering Method Using k NN for Robust Classification

摘要

The correctness and efficiency of data mining results is dependent on data quality. Thus, improving the quality of data holds considerable research significance. This work presents a high quality data selection method known as suspected data noise filtering. Firstly, we propose a data quality evaluation measure named the NN-kNN measure. Then, a suspected noise filtering algorithm based on NN-kNN is proposed. Experiments are carried out to verify the proposed quality evaluation measure and noise filtering algorithm. The experimental results show that NN- k NN can evaluate data quality well. High quality samples selected by the proposed algorithm can generate higher classification accuracy and robust classification results.