An improved elastic net clustering algorithm with dynamic parameter strategy

Clustering is a typical and important method to discover new structures and knowledge from data sets. However, due to the difficulty of achieving high-quality clustering solutions for diverse types of data sets especially for large-scale data sets, and the high computational complexity, how to conduct effective data mining has become a challenge. In order to tackle these issues, we propose an improved elastic net clustering algorithm with dynamic parameter strategy (IENDP). First, we design a novel energy function according to the aim of clustering, which can help the network well distinguish the probability distribution of the data points affiliated with a specific cluster and obtain better clustering solutions, especially for high-dimensional and large-scale problems. Second, a dynamic parameter strategy is introduced into the energy function, which can make the network have higher space searching ability, speed up the expansion and convergence process, and decrease the sensibility of the parameters. The new energy function with dynamic parameter strategy can significantly reduce the impact of the internal structure of the dataset, identify clusters of different sizes, shapes, and densities, and obtain higher clustering quality. Moreover, the proposed IENDP algorithm is a self-organizing and self-learning algorithm that does not require manual guidance and training. Theoretical analysis and experimental results on a large number of synthetic and real-world datasets show that the proposed IENDP can effectively improve the clustering quality with low computational and time complexity, and has superior performance than some classical and state-of-the-art clustering algorithms.