Abstract: In response to the prevalent issue of cost overruns in construction projects and the inherent limitations of conventional cost prediction methods in accounting for project dynamics, this study establishes a machine learning-based dynamic cost prediction model by incorporating critical dynamic influencing factors such as material price fluctuations, design change frequency, and labor productivity variations. A systematic comparison was conducted on the predictive performance of four algorithms: multiple linear regression, support vector machines, random forests, and long short-term memory (LSTM) networks. The empirical results, derived from 50 historical project datasets, demonstrate that machine learning models offer superior predictive accuracy. Among these, random forests and LSTM networks exhibit the highest performance, with the LSTM model showing a marked proficiency in capturing temporal cost characteristics. This research delivers a practical decision-support tool for cost control throughout the project lifecycle.