Abstract
II
Research on High Dimensional Data Clustering Algorithm
Based on Deep Learning
ABSTRACT
With the rapid development of information technology, the scale and dimensions of data
are constantly growing, and the data presents high-dimensional characteristics.Clustering is
the most commonly used method for data analysis. However, due to the large number of
irrelevant attributes, sparse distribution and computational complexity in high-dimensional
data, traditional clustering algorithms do not perform well on high-dimensional data. To solve
the high-dimensional data clustering problem, subspace clustering method is an effective
solution, which transforms the high-dimensional feature space into the low-dimensional
feature space for clustering. It can be implemented by methods such as principal component
analysis (PCA), sparse subspace clustering algorithm (SSC) and low rank representation
algorithm (LRR).
Although these subspace clustering has achieved good results, but these data
representations learned through shallow models may not capture the complex potential
structure of high-dimensional data; secondly, the entire data is used as a dictionary to learn
features, and it is difficult to handle large-scale data sets. Deep learning is considered to be an
effective means of solving these problems due to its excellent feature learning ability and
rapid reasoning ability.
In view of the above problems, we studied subspace clustering algorithm, autoencoder,
and then proposed an Cascade Subspace Clustering Based on Local Structure
Preservation(ICSC).ICSC algorithm provides a research idea for minimizing the difference of
sample points distributed in two distance metric spaces, which provides a research idea for
feature learning that requires the entire data set as a dictionary. In addition, ICSC uses a
decoder to align feature data with the original data local structure and capture the underlying
structure of high-dimensional data. We conducted related experiments on the ICSC algorithm,
evaluated some of the parameters, and found a parameter that optimizes the performance of
the algorithm. Finally, ICSC was compared with other clustering algorithms on multiple data
sets, and three kinds of data were used. The commonly used clustering evaluation indicators
are used to analyze the experimental results. The results show the effectiveness and
superiority of the ICSC algorithm.
万方数据
剩余41页未读，继续阅读
资源评论
