This work aims to improve security and prevent fraud in blockchain-based healthcare systems by applying machine learning algorithms. We used medical data from sensors and blockchain transactions to apply seven classification methods: logistic regression, decision tree, KNN, SVM, and random forest. Some datasets include human vital signs and Ethereum fraud transactions. The first data preprocessing step included cleaning, normalization, and feature selection. Comparing all the models, random forest was the most accurate as well as the quickest one. The results suggested that our system successfully detected the tainted transactions, and XGBoost turned out to provide the highest level of overall accuracy, equal to 97.92%. It raises healthcare data confidence and safety, is useful for fraud detection, but requires some modifications to be applied in other spheres.