Objective: This paper aims to propose a big data-driven adaptive machine learning algorithm for dynamic model optimization and validate its performance on large-scale datasets. Methods: By incorporating an adaptive optimization mechanism that dynamically adjusts learning rates and regularization parameters in real-time, we enhance the predictive accuracy and adaptability of machine learning algorithms in big data environments. Comparative experiments were conducted using the UCI Adult dataset and Kaggle Titanic dataset, with Support Vector Machines (SVM) and Random Forests (RF) serving as baseline models. Results: Experimental results demonstrate that the adaptive optimization-based algorithm significantly outperforms traditional models across metrics including accuracy (Adult: 92.5%, Titanic: 89.3%), precision, recall, and F1-Score. Notably, the F1-Score improved by 8.6% on the Titanic dataset. Conclusion: Adaptive optimization models effectively enhance the performance of machine learning algorithms in big data environments, demonstrating strong generalization capabilities and promising application prospects.
Pei Shen (Thu,) studied this question.