Improvement on Agglomerative Hierarchical Clustering Algorithm Based on Tree Data Structure with Bidirectional Approach

Abstract

Hierarchical clustering algorithms take an input of pair wise data-item similarities and output a hierarchy of the data-items. This paper presents bi-directional agglomerative hierarchical clustering algorithm to create a bottom-up hierarchy, by iteratively merging the closest pair of data-items into one cluster. The result is a rooted AVL tree. The n leafs correspond to input data-items that need to n/2 or n/2+1 steps to merge into one cluster, correspond to groupings of items in coarser granularities climbing towards the root. As observed from the time complexity and number of steps needed to cluster all data points into one cluster perspective, the performance of the bi-directional agglomerative algorithm using tree data structure is better than the current agglomerative algorithms. Analysis on the experimental results indicates that the improved algorithm has a higher efficiency than previous methods.