An Evolutionary Multi-Layer Extreme Learning Machine for Data Clustering Problems

Clustering technology such as K-means is a critical method in data analysis, and it occupies an even more important place with the explosion of information. Due to the complicated distribution of data in the original feature space, it is difficult to find suitable clusters of the data. To address this problem, Existing extreme learning machines (ELM) based clustering methods construct an embedding space to reveal the clusters more easily. In ELM and K-means, some parameters are randomly assigned, so it may lead to unstable or non-optimal results. In this paper, an evolutionary extreme learning machine method (LBH-MLELM) based on a black hole algorithm is proposed, in which a more effective embedding space is constructed by a multi-layer extreme learning machine, and the final results are subsequently derived by K-means algorithm. What's more, by optimizing not only the input weights, hidden biases of ELM, but also the initial center of K-means, the accuracy and robustness of clustering are improved. The experimental results on several public datasets demonstrated the effectiveness of the proposed method, and further analyzed the properties of the proposed method.