In this work, we sought to predict the necrosis score, a surrogate of radiotherapy treatment outcome for sarcoma patients, using the longitudinal diffusion MRI data. Over three hundred features were extracted from the longitudinal diffusion data on twenty sarcoma patients. Minimum redundancy maximum relevance method with cross-validation was used to select the most relevant and stable features. Logistic regression, support vector machine and adaptive boosting were implemented to predict the necrosis score. AUC of 0.76 was achieved when using SVM with features from all three imaging time points. Features from before the treatment time point had better predictive power than data in the middle or after the treatment.