This paper combines multi-granularity cognitive learning and feature weighting information to construct a robust method for generating weighted granular-balls, thus providing a better multi-granularity representation for high-dimensional data.Based on the multi-granularity representation of high-dimensional data, a weighted clustering method that is more adaptive and has better clustering effect is constructed.The effectiveness and accuracy of our proposed method for high-dimensional data characterization is confirmed after extensive testing and comparison on real datasets. This provides a new and effective way to solve the problem of high-dimensional data clustering.

These program mainly containing: data: The folder where the dataset is stored; W_GBC_V1.py: The main function entry point for W-GBC. W_GBC_split_by_k: The specific implementation process of constructing weighted granules and performing weighted clustering in W-GBC. wkmeans_no_random.py: The process of obtaining weights based on an improved version of wkmeans.

The data folder provides the data used in the experiment. The data and labels are in csv format, with the data file named "_X_1_0.csv" and the labels file named "_Y_1_0.csv".

The parameters input to the main function in WGBC_Test_V1.py are as follows: keys: The name of the dataset. K: The number of clusters. The final output includes the evaluation metrics RI, NMI, and ACC.