人工智能教学资料 Project Progress Report.pdfVIP

Project Progress Report Group Members 黄文楷 黄少墅 谢敏 查娟 李正 Introduction We intent to realize automatic divisive hierarchical clustering method, DIVFRP for short. Its basic idea is to bipartition clusters repeatedly with a novel dissimilarity measure based on furthest reference points. A sliding average of sum-of-error is employed to estimate the cluster number preliminarily, and the optimum number of clusters is achieved after spurious clusters identified. We have already implemented the full algorithm. Next step, our task is to test and evaluate DIVFRP through practical data sets. Algorithm Evaluation Advantage: 1. Doesn’t require any user-specified parameter. Most traditional clustering methods, such as K-means, DBScan, require some user-specified parameters. Generally, however, the required parameters are hard to specify and influence heavily to the performance of algorithm. 2. Lower computational cost. DIVFRP employs furthest reference points to bipartition a cluster optimally, and does not need to consider the complete enumeration of all possible bipartitions. For a data set with N objects, the computational cost of DIVFRP is lower than that of single-linkage O (N2log N). Because in DIVFRP, totally there are N -1 bipartitions after every object becomes a cluster, and the computational cost of each bipartition is O (nilog ni), where ni is the object number of a cluster to bipartition and it is less than N except the first bipartition in which ni is N. 3. Support the requirement of the number of clusters as an input from users. Some clustering methods, such as K-means, need number of clusters as an input from users. Well, DIVFRP can help to find the best number of clusters. 4. Robust to outliers. The presence of outliers may deteriorate the result of a clustering method. For this kind of clustering methods, some outlier detection mechanisms can be employed to remove the ou