Abstract:Clustering is an important data analysis technique. However, due to the diversity of datasets, each clustering algorithm is unable to produce satisfactory results on some particular datasets. In this paper, we propose a clustering optimization method called HIAC (Highly Improving the Accuracy of Clustering algorithms). By introducing gravitation, HIAC forces objects in the dataset to move towards similar objects, making the ameliorated dataset more friendly to clustering algorithms (i. e., clustering algorithms can produce more accurate results on the ameliorated dataset). HIAC is independent of clustering principle, so it can optimize different clustering algorithms. In contrast to other similar methods, HIAC is the first to adopt the selective-addition mechanism, i.e., only adding gravitation between valid-neighbors, to avoid dissimilar objects approaching each other. In order to identify valid-neighbors from neighbors, HIAC introduces a decision graph, from which the naked eye can observe a clear division threshold. Additionally, the decision graph can assist HIAC in reducing the negative effects that improper parameter values have on optimization. We conducted numerous experiments to test HIAC. Experiment results show that HIAC can effectively ameliorate high-dimensional datasets, Gaussian datasets, shape datasets, the datasets with outliers, and overlapping datasets. HIAC greatly improves the accuracy of clustering algorithms, and its improvement rates are far higher than that of similar methods. The average improvement rate is as high as 253.6% (except maximum and minimum). Moreover, its runtime is significantly shorter than that of most similar methods. More importantly, with different parameter values, the advantages of HIAC over similar methods are always maintained. The code of HIAC is available at https://github.com/qiqi12/HIAC.

A clustering heuristic to improve a derivative-free algorithm for nonsmooth optimization

A New Diagonal Quasi-Newton Updating Method With Scaled Forward Finite Differences Directional Derivative for Unconstrained Optimization

Enhancing Optimization Performance: A Novel Hybridization of Gaussian Crunching Search and Powell's Method for Derivative-Free Optimization

Scalable Subspace Methods for Derivative-Free Nonlinear Least-Squares Optimization

Improvement of the Nelder-Mead method using Direct Inversion in Iterative Subspace

A descent subgradient method using Mifflin line search for nonsmooth nonconvex optimization

A descent subgradient method using Mifflin's line search for nonsmooth nonconvex optimization

An efficient constrained global optimization algorithm with a clustering-assisted multiobjective infill criterion using Gaussian process regression for expensive problems

An Efficient Smoothing Proximal Gradient Algorithm for Convex Clustering

A Riemannian Optimization Approach to Clustering Problems

A Descent Method for Nonsmooth Multiobjective Optimization in Hilbert Spaces

Iterative Optimization and Simplification of Hierarchical Clusterings

Derivative-Free Optimization via Finite Difference Approximation: An Experimental Study

Expected decrease for derivative-free algorithms using random subspaces

How to improve the accuracy of clustering algorithms

An effective subgradient algorithm via Mifflin's line search for nonsmooth nonconvex multiobjective optimization

A semismooth conjugate gradients method – theoretical analysis

Combining Gradient Information and Primitive Directions for High-Performance Mixed-Integer Optimization

Improved Algorithms Based on Trust Region Framework for Solving Unconstrained Derivative Free Optimization Problems

Exploiting Negative Curvature in Conjunction with Adaptive Sampling: Theoretical Results and a Practical Algorithm