Optimisation Research on K-Means Clustering Algorithm Based on Principal Component Analysis and Percentile Improvement

The traditional K-means algorithm often leads to unstable clustering quality due to the randomness of the initial clustering center selection and tends to fall into suboptimal solutions when dealing with large-scale datasets. In this paper, we study and propose an improved K-means clustering algorithm based on principal component analysis (PCA) and percentile method to solve the problem of randomness of the traditional K-means algorithm in the selection of initial clustering centers. By introducing the PCA dimensionality reduction technique, the high-dimensional data are mapped to a low-dimensional space, thus preserving the main feature information of the data and simplifying the computational complexity. Subsequently, the percentile method is used to select the initial clustering centers on the dimensionality-reduced data, which makes the initial centroids more representative and decentralized, and thus improving the accuracy and consistency of the clustering results. The improved K-means algorithm was validated on several datasets, including the KDD Cup 1999 dataset, the Wine Quality dataset, and the Breast Cancer Wisconsin dataset, and demonstrated that the improved K-means algorithm has significantly improved both the clustering accuracy and the operational efficiency.