As machine learning continues to achieve remarkable success, data has become a critical resource, making public data pricing an essential research topic. This paper explores key challenges in public data pricing, focusing on dataset pricing, label pricing, revenue allocation in collaborative machine learning, and machine learning model pricing, and proposes corresponding solutions. Firstly, we discuss current dataset pricing methods, assessing their applicability and limitations. Next, a dynamic pricing model for labels is introduced, based on task difficulty and sustained performance, aiming to set fair prices for labels of varying quality. For revenue allocation in collaborative machine learning, a multi-level distribution model is proposed to ensure equitable income distribution among participants according to their contributions. Lastly, we present a Machine Learning-Driven Pricing Model (MLDPM) for pricing machine learning models, which adjusts dynamically based on the utility derived from both data and models, and validate its effectiveness through experiments. This study provides theoretical insights and practical frameworks to address key challenges in public data pricing, contributing to the development of efficient pricing mechanisms in machine learning.