Soft Tissue Tumors (STT) are a form of sarcoma found in tissues that connect, support, and surround body structures. Because of their shallow frequency in the body and their great diversity, they appear to be heterogeneous when observed through Magnetic Resonance Imaging (MRI). They are easily confused with other diseases such as fibroadenoma mammae, lymphadenopathy, and struma nodosa, and these diagnostic errors have a considerable detrimental effect on the medical treatment process of patients. Researchers have proposed several machine learning models to classify tumors, but none have adequately addressed this misdiagnosis problem. Also, similar studies that have proposed models for evaluation of such tumors mostly do not consider the heterogeneity and the size of the data. Therefore, we propose a machine learning-based approach which combines a new technique of preprocessing the data for features transformation, resampling techniques to eliminate the bias and the deviation of instability and performing classifier tests based on the Support Vector Machine (SVM) and Decision Tree (DT) algorithms. The tests carried out on dataset collected in Nur Hidayah Hospital of Yogyakarta in Indonesia show a great improvement compared to previous studies. These results confirm that machine learning methods could provide efficient and effective tools to reinforce the automatic decision-making processes of STT diagnostics.