Research on Data Analysis of Traditional Chinese Medicine with Improved Differential Evolution Clustering Algorithm

Honglei Zhu, Yingying Zhao, Xueyun Wang, Yulong XuJournal of Healthcare Engineering 2021 · DOI

TL;DR

This paper introduces an improved Differential Evolution (DE) clustering algorithm designed to overcome the limitations of traditional K-Means, particularly its tendency to get stuck in local optima. By enhancing the DE algorithm's mutation strategy and incorporating a novel diversity indicator, the proposed method achieves better global optimization and faster convergence, demonstrating its effectiveness in both benchmark datasets and a real-world application in Traditional Chinese Medicine (TCM) drug analysis.

What Problem Does This Paper Address?

Clustering is a valuable tool for analyzing complex medical data, especially in Traditional Chinese Medicine (TCM) for syndrome differentiation and drug analysis. However, the widely used K-Means algorithm suffers from a critical drawback: its performance is highly dependent on the initial selection of cluster centers. This often leads to suboptimal clustering results, trapping the algorithm in local optima rather than finding the true global optimum. While other optimization algorithms have been combined with K-Means, this paper aims to improve upon these by proposing a novel approach that leverages an enhanced Differential Evolution (DE) algorithm to more effectively find optimal cluster centers.

Methods

The core of this paper is an Improved Differential Evolution-based K-Means Clustering Algorithm. Here's a breakdown of the key components:

  • Enhanced Differential Evolution (DE): The authors modify the standard DE algorithm, a population-based metaheuristic known for its robustness and global search capabilities.
  • Population Initialization: Unlike traditional K-Means, the initial clustering centers are randomly generated, providing a larger search space.
  • Population Diversity-Based Double-Mutation Operation: A novel approach is introduced to dynamically adjust the mutation strategy based on the current population's diversity.
  • Diversity Indicator: A new indicator, $\lambda(g)$, is proposed to measure population diversity by considering the distance of individuals to the central individual. This indicator helps assess whether the population is too clustered or too spread out.
  • Dynamic Mutation Strategy: In the early stages of evolution (high diversity), the DE/best/1 strategy is used to guide the search towards the optimal individual, promoting faster convergence. As diversity decreases (late stages), the DE/rand/1 strategy is employed to explore more broadly and avoid local optima.
  • Fitness Function: The sum of within-class distances (WCD) is used as the fitness function, aiming to minimize the distances between data points and their assigned cluster centers.
  • Integration with K-Means: The improved DE algorithm is used to optimize the selection and location of the clustering centers, effectively replacing the traditional K-Means center update mechanism. This allows the algorithm to continuously optimize cluster centers during the evolution process.

Datasets

The paper evaluates its proposed algorithm using:

  • Three UCI Standard Datasets:

    • Iris
    • Wine
    • Zoo These are standard benchmarks for evaluating clustering algorithms.
  • Traditional Chinese Medicine (TCM) Drug Data: A dataset comprising 270 data points from 39 kinds of TCM works and 16 literatures, focusing on the treatment of diffuse pulmonary interstitial disease. This dataset is used to demonstrate the algorithm's applicability to real-world medical data and its ability to uncover clinically relevant patterns.

Key Insights and Ideas

  • Dynamic Mutation Strategy is Key: The most significant insight is the development of a dynamic mutation strategy for DE that adapts to population diversity. This allows the algorithm to balance exploration (avoiding local optima) and exploitation (converging to the best solution) more effectively throughout the optimization process.
  • Diversity Indicator for Better Control: The proposed $\lambda(g)$ indicator provides a more nuanced way to assess population diversity compared to simple variance, enabling more precise control over the mutation strategy.
  • Global Optimization for Clustering: The paper strongly advocates for using evolutionary algorithms like DE to overcome the inherent local optima problem of K-Means, leading to more robust and accurate clustering results.
  • TCM Drug Analysis Potential: The successful application to TCM drug data suggests that this improved clustering method can uncover meaningful patterns in complex medical datasets, potentially aiding in understanding drug interactions and treatment strategies.

Key Results

  • Superior Global Optimization: Across all three UCI datasets, the proposed algorithm consistently outperformed K-Means, K-Means++, and the standard DE-K-Means algorithm in terms of global optimization, achieving better clustering effects.
  • Improved Convergence Speed: The proposed algorithm demonstrated a faster convergence speed compared to the standard DE-K-Means algorithm, reaching optimal solutions earlier.
  • TCM Data Consistency: The clustering results for TCM drug data were found to be consistent with established TCM theory, validating the algorithm's practical utility in medical applications. For instance, the identified drug classes align with TCM principles for treating pulmonary diseases, suggesting the algorithm can identify clinically relevant groupings.
  • Stability and Accuracy: The paper highlights that the proposed algorithm offers improved stability and accuracy of clustering results compared to K-Means and K-Means++.

Limitations and Future Work

  • Parameter Tuning: While the paper specifies parameters like mutation factor (F) and crossover probability (CR), the optimal settings for these parameters might still require fine-tuning for different datasets.
  • Computational Cost: Evolutionary algorithms, including DE, can be computationally intensive, especially for very large datasets. While the proposed algorithm shows improved convergence speed, further optimization for extremely large-scale data might be necessary.
  • Future Work: The authors suggest applying the algorithm to analyze more TCM data and exploring its potential in other medical domains. Further research could also focus on developing more sophisticated diversity indicators or adaptive parameter control mechanisms within the DE framework.

Summary updated

Abstract

Medical data analysis is an important part of intelligent medicine, and clustering analysis is a commonly used method for data analysis of Traditional Chinese Medicine (TCM); however, the classical <i>K</i>-Means algorithm is greatly affected by the selection of initial clustering center, which is easy to fall into the local optimal solution. To avoid this problem, an improved differential evolution clustering algorithm is proposed in this paper. The proposed algorithm selects the initial clustering center randomly, optimizes and locates the clustering center in the process of evolution iteration, and improves the mutation mode of differential evolution to enhance the overall optimization ability, so that the clustering effect can reach the global optimization as far as possible. Three University of California, Irvine (UCI), data sets are selected to compare the clustering effect of the classical <i>K</i>-Means algorithm, the standard DE-<i>K</i>-Means algorithm, the <i>K</i>-Means++ algorithm, and the proposed algorithm. The experimental results show that, in terms of global optimization, the proposed algorithm is obviously superior to the other three algorithms, and in terms of convergence speed, the proposed algorithm is better than DE-<i>K</i>-Means algorithm. Finally, the proposed algorithm is applied to analyze the drug data of Traditional Chinese Medicine in the treatment of pulmonary diseases, and the analysis results are consistent with the theory of Traditional Chinese Medicine.
BibTeX
@article{zhu2021research,
  author = {Honglei Zhu and Yingying Zhao and Xueyun Wang and Yulong Xu},
  title = {{Research on Data Analysis of Traditional Chinese Medicine with Improved Differential Evolution Clustering Algorithm}},
  journal = {Journal of Healthcare Engineering},
  year = {2021},
  doi = {10.1155/2021/4468741},
}
Figure