In this study, we considered projects developing service systems using machine learning (ML) techniques. Because of the extensive adoption of ML techniques across various fields, there exists a body of reusable knowledge from ML projects that can be leveraged to execute such projects successfully. However, this knowledge has not been explicitly documented as reusable knowledge, such as best practices or patterns. Therefore, in this study, we propose a method for identifying warning signs in ML projects, known as “bad smells,” and developing anti-patterns from them. Through an experiment, we identified 18 bad smells, referencing existing best practices as solutions, and developed 14 anti-patterns for ML projects. Based on the results of a survey on the identified bad smells, we confirmed that the developed anti-patterns could be broadly applied in ML project practices.