Application of Wave Cluster Algorithm Based on Kernel Optimization in Pattern Recognition of High-Frequency Financial Data

High-frequency financial data has the characteristics of high nonlinearity, non-stationarity and noise, which brings great challenges to traditional data mining algorithms. In this study, we propose a Wave Cluster algorithm based on kernel function optimization, which aims to improve its performance in pattern recognition of high-frequency financial data. The Wave Cluster algorithm has attracted much attention for its robustness to noise and multiresolution analysis capability, but it still has limitations when dealing with complex nonlinear patterns. In this paper, kernel-based functions are introduced to map the original high-dimensional financial data to a higher-dimensional feature space. This nonlinear transformation can effectively capture the potentially complex relationships in the data. Subsequently, the modified Wave Cluster algorithm is applied in the transformed feature space, and high-frequency trading data from the China stock market, including price, trading volume and order book information, are used in the empirical study. Experimental results show that compared with traditional methods, the proposed algorithm achieves significant improvement in the tasks of identifying market microstructure, detecting abnormal trading patterns and predicting short-term price movements. Especially during periods of high volatility, the algorithm shows greater stability and accuracy.