Wait or Reset Gas Price?: A Machine Learning-based Prediction Model for Ethereum Transactions' Waiting Time

Abstract

The gas mechanism on the Ethereum blockchain attempts to set charges for every smart contract operation in order to prevent infinite control of computational resources by the executing transactions. To satisfy this requirement, users need to specify in their transactions how much gas fees (in terms of gas limit and gas price) they would like to pay for. In essence, these gas fees are paid to the miners in return for their computational services. Naturally, miners tend to maximize their profits by considering the transactions with higher gas fees, and as a result, the transactions with lower gas fees remain in the waiting pool for a long time. This paper proposes a machine learning-based approach to predict whether a transaction with offered gas fees is likely to be included in the blockchain within the expected time or not. Such prior prediction of transactions' waiting time definitely assists users to reset their gas fees accordingly. The proposed model is evaluated on nearly one million real transactions from Ethereum mainnet, and the experimental results demonstrate a better performance than the existing one in the literature, with an achievement of 90.18% accuracy and 0.897 F1-score when the model is trained with Random Forest on the dataset balanced with SMOTETomek.