PROJECT TITLE :

Adaptive Noise Immune Cluster Ensemble Using Affinity Propagation

ABSTRACT:

Cluster ensemble is one of the most branches within the ensemble learning space which is a crucial analysis focus in recent years. The objective of cluster ensemble is to combine multiple clustering solutions in a very appropriate way to enhance the standard of the clustering result. In this paper, we style a new noise immune cluster ensemble framework named as $AP^2CE$ to tackle the challenges raised by noisy datasets. $AP^2CE$ not solely takes advantage of the affinity propagation algorithm (AP) and also the normalized cut algorithm (Ncut), but also possesses the characteristics of cluster ensemble. Compared with ancient cluster ensemble approaches, $AP^2CE$ is characterized by many properties. ($1$ ) It adopts multiple distance functions rather than a single Euclidean distance perform to avoid the noise related to the gap function. ( $a pair of$ ) $AP^2CE$ applies AP to prune noisy attributes and generate a group of recent datasets within the subspaces consists of representative attributes obtained by AP. ( $three$ ) It avoids the express specification of the number of clusters. ($4$ ) $AP^2CE$ adopts the normalized cut algorithm as the consensus perform to partition the consensus matrix and get the final result. So as to enhance the performance of $AP^2CE$, the adaptive $AP^2CE$ is intended, that makes use of an adaptive process to optimize a newly designed objective perform. The experiments on each synthetic and real datasets show that ($1$ ) $AP^2CE$ works well on most of the datasets, in explicit the noisy datasets; ($2$ ) $AP^2CE$ could be a better alternative for many of the datasets compared with other cluster ensemble approaches; (