An Attention-GRU Based Gas Price Prediction Model for Ethereum Transactions

Each operation on the Ethereum network requires a corresponding gas cost, and the gas price is customized by the user who uses it. Under normal circumstances, the shorter the user’s requirement for the transaction completion time, the higher the gas price should be set. However, users cannot set a reasonable gas price every time, which will cause operation failure or gas waste. Therefore, predicting an affordable gas price is of great significance. We propose a gas price prediction method based on deep learning to find the lowest transaction gas price in the next block and help users save transaction costs. Based on previous research, we obtained influence factors of block gas upper limit, transaction gas upper limit and mining difficulty that cause gas price fluctuations, combined with the attention mechanism to obtain the weight of the influence factor, we use the gated recurrent unit (GRU) neural network to determine the gas price prediction. Extensive experiments based on 2000 actual blocks show that the proposed gas price prediction method has high accuracy and reliable performance, which saves users a large amount of transaction costs; because the appropriate gas price can be predicted, the success rate of users’ transactions increases by 22.9%.