In recent years, the usage of cryptocurrencies and the acceptance towards the cryptocurrency practices in the financial systems has increased throughout the world. The cryptocurrencies that are developed based on blockchain technology are believed to be the safest as it incorporates the properties like decentralization, immutability and consensus mechanism. However, activities such as 51% attack, double spending, phishing, vulnerable smart contracts, Sybil attacks can reduce the durability of the cryptocurrencies. According to Chainalysis Inc., in 2021, the value of the fraudulent transaction amounted to a worth of 14 billion dollars worldwide. Since cryptocurrencies are decentralized, the lack of centralized monitoring mechanism opens a path for the intruders to perform illicit transactions. To overcome this issue and to ensure the sustainability of blockchain based cryptocurrencies, this research work presents machine learning model to detect fraudulent transactions. To train the model, Ethereum Fraud Detection Dataset is used to train the model. However, the dataset is imbalanced with the legit transaction and the illicit transaction. To balance the dataset SMOTE and ADASYN preprocessing techniques are used. The fine-tuned dataset is trained with five different machine learning algorithms such as Support Vector Machine (SVM), Naïve Bayes, KNN algorithm, Random Forest and Logistic Regression. To evaluate the performance of the proposed model, accuracy, precision, recall and F1-Score of each model is assessed individually. Among these machine learning model, Random Forest attained a maximum accuracy of 97% and the second highest accuracy is attained by KNN Model. Algorithms such as Logistic Regression, SVM and Naïve Bayes model attained 78.8%, 78% and 35% respectively.