This repository provides a curated list of deanonymized transaction-level features originally released in anonymized form as part of the Elliptic Bitcoin Dataset.
The feature semantics were recovered through systematic reverse engineering and validation against independently computed blockchain statistics.
The primary purpose of this repository is to document the recovered meanings of Elliptic features and to support reproducible research in Bitcoin transaction analysis and forensic machine learning.
The Elliptic Bitcoin Dataset (https://www.kaggle.com/datasets/ellipticco/elliptic-data-set) is widely used for benchmarking machine learning models for illicit transaction detection. However, the original dataset does not publish feature definitions, which limits:
- interpretability of trained models,
- reproducibility of experimental results,
- applicability of models to real-world forensic investigations.
The feature meanings documented here were identified and validated as part of an academic study that also demonstrates multiple forms of data leakage present in the original Elliptic dataset.
The feature definitions and semantic mappings documented in this repository are not part of the original Elliptic dataset and were independently reconstructed.
If you use any feature definitions, interpretations, or mappings from this repository in academic work, please cite the following paper:
Šafář, M., Pluskal, J., Veselý, V., Ryšavý, O. (2026). The enemy of reproducibility is opacity: What's inside the Elliptic bitcoin dataset (and why it is wrong). Forensic Science International: Digital Investigation, 57, 302124. https://doi.org/10.1016/j.fsidi.2026.302124
BibTeX
@article{SAFAR2026302124,
title = {The enemy of reproducibility is opacity: What's inside the Elliptic bitcoin dataset (and why it is wrong)},
journal = {Forensic Science International: Digital Investigation},
volume = {57},
pages = {302124},
year = {2026},
issn = {2666-2817},
doi = {10.1016/j.fsidi.2026.302124},
url = {https://www.sciencedirect.com/science/article/pii/S2666281726000818},
author = {Miroslav Šafář and Jan Pluskal and Vladimír Veselý and Ondřej Ryšavý}
}
This repository does not redistribute the Elliptic dataset or any derived dataset, in accordance with the Elliptic CC BY-NC-ND 4.0 license. It only provides feature semantics and documentation, which were independently derived through analysis of publicly available blockchain data.
The reconstructed information is provided for research, verification, and reproducibility purposes. Access to and use of the original Elliptic dataset remain subject to the terms imposed by its original authors and distributors.
When using information from this repository, please cite:
Šafář, M., Pluskal, J., Veselý, V., Ryšavý, O. (2026). The enemy of reproducibility is opacity: What's inside the Elliptic bitcoin dataset (and why it is wrong). Forensic Science International: Digital Investigation, 57, 302124. https://doi.org/10.1016/j.fsidi.2026.302124
Important: The feature meanings below were recovered by Šafář et al. and are not provided by Elliptic.
Please cite the paper when reproducing or explaining these feature definitions.
| Feature | Description |
|---|---|
| Feature 1 | Time Step |
| Feature 2 | Transaction Volume |
| Feature 3 | Fees |
| Feature 4 | - |
| Feature 5 | Number of inputs |
| Feature 6 | Number of outputs |
| Feature 7 | Number of unique input addresses |
| Feature 8 | - |
| Feature 9 | Minimum of input values |
| Feature 10 | Maximum of input values |
| Feature 11 | Standard deviation of input values |
| Feature 12 | Mean of input values |
| Feature 13 | Pearson correlation between input values and indices |
| Feature 14 | Spearman correlation between input values and indices |
| Feature 15 | Number of unique output addresses |
| Feature 16 | - |
| Feature 17 | Minimum of output values |
| Feature 18 | Maximum of output values |
| Feature 19 | Standard deviation of output values |
| Feature 20 | Mean of output values |
| Feature 21 | Pearson correlation between output values and indices |
| Feature 22 | Spearman correlation between output values and indices |
All address-level features (i.e., features computed from input/output addresses and then aggregated into transaction features 23–94) are computed relative to a single fixed reference point in time: Bitcoin block height 575,059.
In other words, balances, counts, and lifetimes reflect the state/history of each address as observed up to block 575,059, rather than being evaluated at each transaction’s own time step.
| Feature | Description |
|---|---|
| Feature 23 | Minimum number of received satoshi of input addresses |
| Feature 24 | Maximum number of received satoshi of input addresses |
| Feature 25 | Standard deviation of number of received satoshi of input addresses |
| Feature 26 | Mean number of received satoshi of input addresses |
| Feature 27 | Pearson correlation between number of received satoshi of input addresses and indices |
| Feature 28 | Spearman correlation between number of received satoshi of input addresses and indices |
| Feature 29 | Minimum number of sent BTC of input addresses |
| Feature 30 | Maximum number of sent BTC of input addresses |
| Feature 31 | Standard deviation of number of sent BTC of input addresses |
| Feature 32 | Mean number of sent BTC of input addresses |
| Feature 33 | Pearson correlation between number of sent BTC of input addresses and indices |
| Feature 34 | Spearman correlation between number of sent BTC of input addresses and indices |
| Feature 35 | Minimum available BTC on input addresses |
| Feature 36 | Maximum available BTC on input addresses |
| Feature 37 | Standard deviation of available BTC on input addresses |
| Feature 38 | Mean available BTC on input addresses |
| Feature 39 | Pearson correlation between available BTC on input addresses and indices |
| Feature 40 | Spearman correlation between available BTC on input addresses and indices |
| Feature 41 | Minimum number of incoming transactions of input addresses |
| Feature 42 | Maximum number of incoming transactions of input addresses |
| Feature 43 | Standard deviation of number of incoming transactions of input addresses |
| Feature 44 | Mean number of incoming transactions of input addresses |
| Feature 45 | Pearson correlation between number of incoming transactions of input addresses and indices |
| Feature 46 | Spearman correlation between number of incoming transactions of input addresses and indices |
| Feature 47 | Minimum number of outgoing transactions of input addresses |
| Feature 48 | Maximum number of outgoing transactions of input addresses |
| Feature 49 | Standard deviation of number of outgoing transactions of input addresses |
| Feature 50 | Mean number of outgoing transactions of input addresses |
| Feature 51 | Pearson correlation between number of outgoing transactions of input addresses and indices |
| Feature 52 | Spearman correlation between number of outgoing transactions of input addresses and indices |
| Feature 53 | Minimum lifetime of input addresses |
| Feature 54 | Maximum lifetime of input addresses |
| Feature 55 | Standard deviation of lifetime of input addresses |
| Feature 56 | Mean lifetime of input addresses |
| Feature 57 | Pearson correlation between lifetime of input addresses and indices |
| Feature 58 | Spearman correlation between lifetime of input addresses and indices |
| Feature 59 | Minimum number of received satoshi of output addresses |
| Feature 60 | Maximum number of received satoshi of output addresses |
| Feature 61 | Standard deviation of number of received satoshi of output addresses |
| Feature 62 | Mean number of received satoshi of output addresses |
| Feature 63 | Pearson correlation between number of received satoshi of output addresses and indices |
| Feature 64 | Spearman correlation between number of received satoshi of output addresses and indices |
| Feature 65 | Minimum number of sent BTC of output addresses |
| Feature 66 | Maximum number of sent BTC of output addresses |
| Feature 67 | Standard deviation of number of sent BTC of output addresses |
| Feature 68 | Mean number of sent BTC of output addresses |
| Feature 69 | Pearson correlation between number of sent BTC of output addresses and indices |
| Feature 70 | Spearman correlation between number of sent BTC of output addresses and indices |
| Feature 71 | Minimum available BTC on output addresses |
| Feature 72 | Maximum available BTC on output addresses |
| Feature 73 | Standard deviation of available BTC on output addresses |
| Feature 74 | Mean available BTC on output addresses |
| Feature 75 | Pearson correlation between available BTC on output addresses and indices |
| Feature 76 | Spearman correlation between available BTC on output addresses and indices |
| Feature 77 | Minimum number of incoming transactions of output addresses |
| Feature 78 | Maximum number of incoming transactions of output addresses |
| Feature 79 | Standard deviation of number of incoming transactions of output addresses |
| Feature 80 | Mean number of incoming transactions of output addresses |
| Feature 81 | Pearson correlation between number of incoming transactions of output addresses and indices |
| Feature 82 | Spearman correlation between number of incoming transactions of output addresses and indices |
| Feature 83 | Minimum number of outgoing transactions of output addresses |
| Feature 84 | Maximum number of outgoing transactions of output addresses |
| Feature 85 | Standard deviation of number of outgoing transactions of output addresses |
| Feature 86 | Mean number of outgoing transactions of output addresses |
| Feature 87 | Pearson correlation between number of outgoing transactions of output addresses and indices |
| Feature 88 | Spearman correlation between number of outgoing transactions of output addresses and indices |
| Feature 89 | Minimum lifetime of output addresses |
| Feature 90 | Maximum lifetime of output addresses |
| Feature 91 | Standard deviation of lifetime of output addresses |
| Feature 92 | Mean lifetime of output addresses |
| Feature 93 | Pearson correlation between lifetime of output addresses and indices |
| Feature 94 | Spearman correlation between lifetime of output addresses and indices |
| Feature | Description |
|---|---|
| Feature 95 | Minimum value of input txs |
| Feature 96 | Maximum value of input txs |
| Feature 97 | Standard deviation of value of input txs |
| Feature 98 | Mean value of input txs |
| Feature 99 | Pearson correlation between value of input txs and indices |
| Feature 100 | Spearman correlation between value of input txs and indices |
| Feature 101 | Minimum height of input txs |
| Feature 102 | Maximum height of input txs |
| Feature 103 | Standard deviation of height of input txs |
| Feature 104 | Mean height of input txs |
| Feature 105 | Pearson correlation between height of input txs and indices |
| Feature 106 | Spearman correlation between height of input txs and indices |
| Feature 107 | Minimum fee of input txs |
| Feature 108 | Maximum fee of input txs |
| Feature 109 | Standard deviation of fee of input txs |
| Feature 110 | Mean fee of input txs |
| Feature 111 | Pearson correlation between fee of input txs and indices |
| Feature 112 | Spearman correlation between fee of input txs and indices |
| Feature 113 | - |
| Feature 114 | - |
| Feature 115 | - |
| Feature 116 | - |
| Feature 117 | - |
| Feature 118 | - |
| Feature 119 | Minimum number of inputs of input txs |
| Feature 120 | Maximum number of inputs of input txs |
| Feature 121 | Standard deviation of number of inputs of input txs |
| Feature 122 | Mean number of inputs of input txs |
| Feature 123 | Pearson correlation between number of inputs of input txs and indices |
| Feature 124 | Spearman correlation between number of inputs of input txs and indices |
| Feature 125 | Minimum number of outputs of input txs |
| Feature 126 | Maximum number of outputs of input txs |
| Feature 127 | Standard deviation of number of outputs of input txs |
| Feature 128 | Mean number of outputs of input txs |
| Feature 129 | Pearson correlation between number of outputs of input txs and indices |
| Feature 130 | Spearman correlation between number of outputs of input txs and indices |
| Feature 131 | Minimum value of output txs |
| Feature 132 | Maximum value of output txs |
| Feature 133 | Standard deviation of value of output txs |
| Feature 134 | Mean value of output txs |
| Feature 135 | Pearson correlation between value of output txs and indices |
| Feature 136 | Spearman correlation between value of output txs and indices |
| Feature 137 | Minimum height of output txs |
| Feature 138 | Maximum height of output txs |
| Feature 139 | Standard deviation of height of output txs |
| Feature 140 | Mean height of output txs |
| Feature 141 | Pearson correlation between height of output txs and indices |
| Feature 142 | Spearman correlation between height of output txs and indices |
| Feature 143 | Minimum fee of output txs |
| Feature 144 | Maximum fee of output txs |
| Feature 145 | Standard deviation of fee of output txs |
| Feature 146 | Mean fee of output txs |
| Feature 147 | Pearson correlation between fee of output txs and indices |
| Feature 148 | Spearman correlation between fee of output txs and indices |
| Feature 149 | - |
| Feature 150 | - |
| Feature 151 | - |
| Feature 152 | - |
| Feature 153 | - |
| Feature 154 | - |
| Feature 155 | Minimum number of inputs of output txs |
| Feature 156 | Maximum number of inputs of output txs |
| Feature 157 | Standard deviation of number of inputs of output txs |
| Feature 158 | Mean number of inputs of output txs |
| Feature 159 | Pearson correlation between number of inputs of output txs and indices |
| Feature 160 | Spearman correlation between number of inputs of output txs and indices |
| Feature 161 | Minimum number of outputs of output txs |
| Feature 162 | Maximum number of outputs of output txs |
| Feature 163 | Standard deviation of number of outputs of output txs |
| Feature 164 | Mean number of outputs of output txs |
| Feature 165 | Pearson correlation between number of outputs of output txs and indices |
| Feature 166 | Spearman correlation between number of outputs of output txs and indices |
The recovered feature meanings are based on reverse engineering and empirical validation. While extensive consistency checks were performed, no affiliation with or endorsement by Elliptic is implied.