A fast incomplete data classification method based on representative points and K-nearest neighbors

Machine learning classification methods play an important role in image recognition, quantitative transaction and disease diagnosis. In the classification task, some sample features come with missing data. Classifiers based on complete features perform poorly in these samples. There are two methods to classify such data: train a new classifier and perform data imputation. The former often takes a very much time when the data volume is large, so the latter is generally selected. Knearest neighbor imputation is a classical data imputation algorithm, but it takes a long time. This paper proposes an imputation algorithm based on K-nearest neighbor imputation(KNNI). We call it a Fast-K-nearest neighbor imputation(FKNNI). The method first roughly classifies each sample and selects a smaller subtraining set before K-nearest neighbor imputation. This paper verifies the speed and accuracy of K-nearest neighbor imputation, and Fast- K-nearest neighbor imputation on three medical datasets. It proves that this method has a good effect on rapid disease diagnosis.