关键词: 大肠早癌诊断数据, 计数最近邻算法, 全局密度, K-密度

Abstract: Data classification is an important data mining role in biomedicine.This paper proposes a method to analyze Colorectal Carcinoma diagnosis data based on counting KNN algorithm after analyzing the characteristics of biomedicine data.Though the count-weight-k-nearest neighbours for classification is simple and effective，it doesn’t deal with biomedicine data well.After analyzing the algorithm performance，an novel counting KNN algorithm by index tree and sample density is presented.The new method improves the accuracy of classification by using different algorithms of overall density and K-local density，and also improves efficiency by using a tree structure index.Experiments show that this method outperforms the distance-based voting KNN，and CwKNN.More importantly it is a single method that works for ordinal，nominal or mixed data.

Key words: colorectal carcinoma diagnosis data, KNN by counting, overall density, K-density