Objectives In this study, a machine learning-based teacher education program for classifying the state of matter was developed and implemented for elementary and secondary science teachers and machine learning model for state classification of matter was established. And by analyzing the model evaluation and the experience of science teachers qualitatively, it was attempted to confirm the effectiveness of the program and the perception of science teachers about the application of machine learning to science classes.
 Methods For 31 elementary and middle school science teachers enrolled in the Graduate School of Education at the College of Education located in the central region, a total of three matter classification activities were performed and a decision tree algorithm was applied to the machine learning model. And the effectiveness of the program was confirmed through model performance evaluation such as accuracy and F1-score. In addition, we qualitatively analyzed the thoughts of science teachers on the application of machine learning.
 Results TAs a result of evaluating the performance of the built model during a total of 3 state of matter classification activities, it was determined that the accuracy and F1-score values increased statically, resulting in the educational effect of the Machine Learning-Based Teacher Education Program on the Classifying State of Matter. And as a result of analyzing the data of the study participants, it was revealed that there are cases in which the states of matter are classified through a deductive thinking process based on knowledge. As a result of comparing evaluation indicators of the machine learning models, the macroscopic classification criteria had a greater effect than the microscopic classification criteria. And when the classification results between the machine learning model and the research participants for various examples were compared, in the case of mixtures, the degree of inconsistency increased. In addition, through qualitative research analysis, the thoughts of science teachers on the application of machine learning were derived into nine topics, including ‘experience the process of creating science concepts’, ‘cognitive conflict’, and ‘communication’.
 Conclusions Machine learning should be actively introduced into teacher education as an inquiry-learning method of science education. And educational materials such as textbooks that can properly reflect scientific reasoning processes such as inductive thinking should be developed based on machine learning-based education programs. And it is necessary to discuss from a scientific point of view whether mixtures should be included in the learning of Classifying State of Matter.