The accuracy of software effort estimation is a critical determinant in the effective planning, management, and successful delivery of software projects within predefined budgets and schedules. The vast diversity of machine learning (ML) techniques has fostered extensive comparisons and driven the integration of these methods within software effort estimation. Given their distinct advantages, it has become essential to identify and prioritize the most effective estimation techniques to enhance the efficiency and accuracy of the software development process. This paper is to present a survey to investigate the machine learning models used in SEE models from 2020 to 2025 and summarize their strengths, limitations and datasets. Our findings show that artificial neural networks demonstrate superior performance in SEE. In addition, fine-tuning models, optimizing parameters, utilizing datasets with effective feature selection, and employing appropriate model selection strategies are critical factors that significantly enhance the accuracy and reliability of software effort estimation models.