Autonomous tumor prostheses are extended without the need of a clinic and of a medical supervision. It is necessary to make sure that the patient is not standing before extending these prostheses. This study aims to determine the posture of the patient for expandable tumor prostheses by employing oft-used three machine learning-based classification methods through comparing them all with each other. Patient posture is determined by using accelerometer and gyroscope data from inertial control unit placed in autonomous expandable tumor prosthesis. By using the created dataset, 48 features are extracted. Then, for optimization, with feature selection, the number of features is reduced to 10. The selected features are processed using the decision tree, the k-nearest neighborhood and support vector machine algorithms. These algorithms were compared with each other using machine learning performance parameters. Accuracy, recall, precision and F-score values are calculated and compared. Consequently, support vector machine is determined as the most successful technique. Then, the model is tested on the experimental setup developed within the scope of the study, and the posture is determined. It is found that with this system, in the presence of a load on the prosthesis, it can be accurately detected at a rate of 97.1% (the recall parameter).