Why It's Difficult to Accurately Predict Ethereum Transaction Times
The block timestamp in blockchain systems is defined as when a new block appears in the Latest Block list of Etherscan, not the precise time it is appended to the chain. The processed timestamp from Etherscan may have a small, consistent error due to the…
The block timestamp in blockchain systems is defined as when a new block appears in the Latest Block list of Etherscan, not the precise time it is appended to the chain. The processed timestamp from Etherscan may have a small, consistent error due to the time taken for Etherscan to recognize and display the new block and for data retrieval. This lag is incorporated into our model's predictions.
Etherscan is a widely used Ethereum dashboard, and its recognition of mined transactions is considered authoritative by many tools. The small error in our estimations is thus expected to have minimal practical impact.
All models must contend with timestamp inaccuracies. Many models likely use blockchain-recorded timestamps, which are more readily available but flawed.
The Ethereum blockchain lacks submission timestamps for transactions, necessitating reliance on Etherscan for pending timestamps, which may affect accuracy.
This study uses available data to estimate transaction processing times, but future work should develop more robust data collection methods over longer periods.
Gas prices were categorized into five groups for ease of analysis, given their volatility. Different categorical classifications would also be valid.
Alpha centrality was used to rank prediction models, as it is appropriate for data with long-tailed distributions and large outliers, unlike other methods such as Scott-Knott ESD.
Etherscan is a widely used Ethereum dashboard, and its recognition of mined transactions is considered authoritative by many tools.
Our weighting system for edges in the Alpha centrality calculation prioritizes wins over draws. This choice does not affect the results provided the weighting condition is met.
Data collection from Etherscan minimized requests, potentially missing some transactions and skewing processing time data. This limitation may prevent verification of certain conjectures, such as dependencies between transactions.
A linear regression model was compared against Random Forests and LightGBM, showing similar performance across all price categories, favoring linear regression for its simplicity and interpretability.
A small percentage of transactions were processed extremely quickly, potentially affecting measurement accuracy.
Data from Ethereum between Nov 21, 2019, and Dec 9, 2019, may not generalize to other periods due to Ethereum's changing workload and contextual factors like market conditions and protocol changes.
Gas prices during the study window were analyzed to evaluate transaction representativeness. Despite recent price surges, the study window is considered a reasonable representation of typical Ethereum transaction history.
Future studies should consider different time frames and factors affecting transaction processing times. The study's reliance on Etherscan data may be affected by network conditions like geolocation and congestion.
Findings are based on standard statistical practices. Developers may prioritize different model accuracy metrics, such as median absolute errors for certain price categories.
Michael Pacheco, Software Analysis and Intelligence Lab (SAIL) at Queen’s University, Canada Gustavo A. Oliva, Software Analysis and Intelligence Lab (SAIL) at Queen’s University, Canada Gopi Krishnan Rajbahadur, Centre for Software Excellence at Huawei, Canada Ahmed E. Hassan, Software Analysis and Intelligence Lab (SAIL) at Queen’s University, Canada
This paper is available on arXiv under CC BY-NC-ND 4.0 DEED license.
Based on reporting by hackernoon.com.