Abstract: The paper presents an improved DBSCAN algorithm for bus station cluster, which improves the degree of accuracy by shrinking the searching radius and merging the clusters by the share object. As a result it efficaciously avoids over-segmentation and reduces the number of noise points. Thus, it effectively shields the sensitivity of the input parameter, produces better clustering results, and reduces the influence to the cluster result by density disparity. And the high performance of the origin algorithm is maintained at the same time. Experimental results demonstrate by cluster of Beijing bus station. And the accuracy of cluster increasesby 16%, which indicates that improve algorithm is valid.

Key words: clustering, Density Based Spatital Clustering of Application with Noise(DBSCAN) algorithm, sensitive to input parameter, data mining

摘要： 提出一种适用于公交站点聚类的DBSCAN改进算法，缩小搜索半径ε，从而提高聚类正确度，同时通过共享对象判定连接簇的合并，防止簇的过分割，减少噪声点，有效地屏蔽了算法对输入参数的敏感性，提高聚类结果的质量，减少密度差距对聚类结果的影响。保持DBSCAN算法的高执行效率，并应用在智能公交换乘查询引擎中公交站点聚类，聚类准确率提高了16%，验证了新算法的有效性。

关键词: 聚类, DBSCAN算法, 参数敏感, 数据挖掘

CLC Number: 

• TP391