With the advent of the big data era, the application of high-dimensional data in various fields has become increasingly widespread. However, traditional statistical methods face many challenges when dealing with high-dimensional data, such as overfitting, multicollinearity and computational complexity. In this paper, we explore machine learning-based parameter estimation methods for high-dimensional statistical models, focusing on the application of supervised learning, unsupervised learning and deep learning techniques in high-dimensional data analysis. This research proposes an efficient parameter estimation framework by incorporating deep neural networks (DNNs), self-encoders, and regularization approaches. According to the experimental findings, the machine learning methodology outperforms conventional techniques in terms of resilience and accuracy of parameter estimation, particularly when handling complicated nonlinear connections and data noise. The limits of the current approaches are also examined in this work, along with suggestions for future research aimed at enhancing computing efficiency, enhancing model interpretability, and expanding the model’s use with small sample sizes.