Research on Stock Price Prediction Model based on Clustering-Sliced LDA and Gaussian Process Regression

Multi-factor prediction techniques based on machine learning models are crucial methods in the field of financial quantification. However, these methods often encounter challenges such as curse of dimensionality and unknown structural connections. To address these issues, this paper proposes an adaptive linear discriminant analysis based on clustering-sliced approach, which can be utilized for continuous response variables. It not only resolves the curse of dimensionality but also extracts local information of predictor variables regarding the response variables. For fitting the unknown structural connections between predictor variables and response variables, Gaussian process regression is employed due to the flexibility of kernel functions, enabling it to fit complex connection functions. Furthermore, Gaussian process regression can simultaneously consider factor information and stock price volatility information, and provide uncertainty interval estimation for predicted values. Simulation data analysis demonstrates that the proposed method exhibits smaller mean squared error, absolute error, and relative error compared to some classical quantitative stock price prediction models. Finally, this model is applied to the task of predicting stock price fluctuations, revealing its higher accuracy and robustness compared to the benchmark methods.