Research on Data filling Algorithm Based on Improved k-means and Information Entropy

摘要

Due to human error, equipment failure and other factors, the industrial Internet platform may generate a part of missing data. In order to fill in the missing data, this paper adopts a data filling method based on improved k-means and information entropy. First, we use the mean or mode to pre-fill the missing data. Then, we change the Euclidean distance in the k-means cluster to the Mahalanobis distance to cluster the data; and within the same category, calculate the similarity between each missing data and all complete data. Finally, Combined with the KNN idea, we find the k complete data that are most similar to each missing data, use information entropy to calculate the weight coefficients of the k complete data, and weight the corresponding attributes of the complete data to fill in the missing attributes. Experimental results show that the data filling algorithm in this paper has better filling precision than k-means and KNN algorithms.