Software Effort Estimation (SEE) is a potential factor in improving the management of Agile projects. However, traditional effort estimation in the dynamic Agile environment, such as expert-based methods, often faces challenges due to subjectivity and inconsistency. Recently, Machine Learning (ML) has emerged as a promising approach, learning from past estimations and leveraging historical data to identify patterns and enhance estimation accuracy. This paper presents a Systematic Literature Review (SLR) on ML-based SEE in Agile environments. We categorize ML techniques, compare their performance with traditional methods, and analyze common evaluation metrics. Our findings show that ML approaches can outperform traditional techniques when applied correctly, though challenges remain, including data availability, evaluation inconsistencies, and external factors. These insights highlight research gaps and suggest directions for future studies.