Improved K-means clustering algorithm based on density radius

The invention, which relates to the field of clustering algorithms, discloses an improved K-means clustering algorithm based on density radius so that problems that a local optimal solution exists, the sensitivity to the noises and outliers is high, and the k value selection is not accurate of the existing K-means clustering algorithm can be solved. All sample points are ranked according to the density radius, the sample point with the largest density radius is used as an initial value, the above-mentioned steps are repeated, all the initial points and the category number k are selected, and clustering operation is stated; two centroids at nearest distances are selected among the clustered category centroids, the categories of the two centroids are taken separately and viewed as a dichotomy, a Bayesian score of the dichotomy is calculated, the two categories are combined into one, a Bayesian score after combination is calculated, whether the two categories need to be combined is determined based on the score,