In recent years, cryptocurrencies have been used as a new way to conduct transactions and transfer money among individuals, so the volume of daily transactions in their networks has reached to several billion dollars. The anonymity of users alongside the high security and privacy properties has led many criminals to the cryptocurrency networks to carry out their illegal transactions. However, public access to the blockchain of Bitcoin and many other cryptocurrencies, allows individuals and financial institutions to obtain information about some of these activities. Several approaches, such as investigating financial flow in the blockchain, statistical analysis, and machine learning methods, have been introduced to detect illegal transactions. This paper uses a deep learning model based on a graph convolutional network and multi-layer perceptron to classify Bitcoin transactions based on their applications. We extract several features from the transaction graph, then by doing some preprocessing on our data, we train a model, which is able to predict illicit transactions with an f1-score of 97.09% which outperforms previous approaches to this problem.