{
  "fetched_at": "2026-10-19T09:30:43Z",
  "source_url": "https://www.kaggle.com/competitions/directional-forecasting-in-cryptocurrencies/data",
  "pages": [
    {
      "name": "data-description",
      "content": "Participants will be provided with a dataset containing historical OHLCV data for a specific cryptocurrency, recorded at a frequency of one minute, meaning there is one data point for each minute. The challenge is to predict the direction of the price movement for the next timestamp.\n\nAs historic cryptocurrency prices are not confidential, the public leaderboard targets are publicly available and are provided as part of the competition dataset. Expect to see many people submitting perfect submissions for fun. Accordingly, the active phase public leaderboard for this competition is not meaningful and is only provided as a convenience for anyone who wants to test their code. In order to do the final evaluation of the participants' solution, the code & model needs to be submitted and will go through separate evaluation. This serves as a check that no look ahead is involved. \n\nThe dataset includes the following columns:\n*  **timestamp**: A timestamp for the minute covered by the row.\n*  **open**: The price in USDT at the beginning of the minute.\n*  **high**: The highest price in USDT during the minute\n*  **low**: The lowest price in USDT during the minute.\n*  **close**: The price in USDT at the end of the minute.\n*  **volume**: The number of crypto asset units traded during the minute.\n*  **quote_asset_volume**: The total value of the traded shares in USDT during the minute.\n*  **number_of_trades**: The number of trades that took place during the minute.\n*  **taker_buy_base_volume**: The total amount of the crypto asset that was bought by takers during the minute.\n*  **taker_buy_quote_volume**: The total value in USDT that was spent by takers to buy the crypto asset during the minute.\n*  **target**: The direction of price movement from the current to the next minute. 1 stands for a positive price change and 0 for no or a negative price change.\n\n## Files\n\n*   **train.csv** - The training data set.\n*   **test.csv** - The test data set, used for the leaderboard.\n*   **sample_submission.csv** - An example of how the submission csv file should look like.\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Prizes",
      "content": "**The best submissions get a consideration for a highly selective program with a career track.**",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "We live in a renaissance era for quantitative finance and decentralized systems. We attempt to combine the two, by building predictors of crypto assets. Cryptocurrency markets operate 24/7, with prices that can change rapidly. For traders and investors, predicting the direction of a crypto asset's price in the short term can be the difference between profit and loss. In this competition, your challenge is to predict the direction of a crypto asset's price movement for the next minute based on historical data.\n\nGiven a dataset containing minute-by-minute price information of one crypto asset, you are tasked with building a model that can accurately predict whether the price of a cryptocurrency will move up (1) or stay the same/move down (0) in the next minute.\n\nThis competition offers a unique opportunity to delve into high-frequency trading insights within the volatile and fast-paced cryptocurrency market. It challenges participants to build robust models capable of predicting short-term price movements, making it highly relevant for both crypto enthusiasts and finance professionals. The task encourages innovative feature engineering, the application of cutting-edge machine learning techniques, and provides a realistic simulation of market impact. Moreover, it fosters collaboration within a global community of data scientists, making it an excellent platform for learning, networking, and showcasing your skills.\n\nDreamers Inc is a science and technology think-tank based in Los Angeles, CA. We consist of a dozen engineers and zero bureaucracy, and have worked with clients from all walks of life: hedge funds, small startups, large companies, universities and researchers. We have a particular interest in emergent technology, applying cutting edge machine learning systems to financial markets, and experimenting with decentralized agreements as a new mode of social and financial operation. Always, the goal in mind is to utilize our technological capabilities to expand the reach of artificial intelligence across different industries, and to work on projects in which technological innovation is in harmony with our values. \n\nClients include: NASA • US Air Force • US Department of Education • University of California Los Angeles • Johns Hopkins University Applied Physics Laboratory • University of California Santa Barbara • University of Michigan • Arizona Court of Appeal• British Aerospace Engineering • Ball Aerospace • Adamant Aerospace • Boeing Aerospace\n\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "1. **Submit Replicable Findings**\nAll submissions are expected to be replicable by a skilled operator in order to be considered. Source control for the competition will be provided such that all findings can be tested. This includes models for inference, and code for training, along with any necessary documentation required for replication of your results. \n\n2. **Individual Entries Only**\nThis competition is strictly limited to individual entries. No teams are allowed, and each participant must submit their work independently. Collaboration between participants is not permitted, and all submissions must be the original work of the individual participant.\n\n3. **One Account per Participant**\nEach participant is allowed to compete using only one Kaggle account. Creating multiple accounts to submit additional entries or gain an unfair advantage is strictly prohibited.\n\n4. **No Private Sharing of Code or Data**\nSharing code or data privately outside what is publicly available in the competition forums is not allowed. If you wish to share code, it must be made accessible to all participants by posting it in the forums. Any private sharing of resources will result in disqualification.\n\n5. **Submission Limits**\nParticipants may submit a maximum of 10 entries per day. It is important to manage your submissions wisely, as the competition's evaluation metrics may fluctuate throughout the day based on new data or insights.\n\n6. **Model and Code Submission for Evaluation**\nTo ensure fairness, all participants are required to submit their complete model and code for training and inference as part of the competition. Upon submission, your code will undergo an additional evaluation process to verify its validity. Only submissions that pass this evaluation will be eligible for inclusion on the official leaderboard and for prize consideration.\n\n7. **No \"Look-Ahead\" Bias**\nDue to the public availability of historical cryptocurrency data, participants must ensure that their models do not use any future data to predict past events (\"look-ahead\" bias). Submissions found to be using future data will be disqualified.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "abstract",
      "content": " ",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "Submissions are evaluated using the Macro-Averaged F1 Score, which is the mean of the F1 scores calculated for each class. The F1 score, a measure of a test's accuracy, is calculated using the formula:\n\n$$\nF1 = 2 \\times \\frac{Precision \\times Recall}{Precision + Recall}\n$$​\n \nWhere:\n- Precision is the ratio of correctly predicted positive observations to the total predicted positives.\n- Recall is the ratio of correctly predicted positive observations to all observations in the actual class.\n\n\nThe Macro-Averaged F1 Score is then computed as:\n\n​$$ F1_{macro} = \\frac{1}{C} \\sum_{c=1}^{C} F1_c $$\n \nWhere:\n- C is the total number of classes (in this case, 2: up or not up).\n- F1_c is the F1 score for class c.\n\nThis evaluation metric ensures that the model performs well across both classes, balancing the precision and recall to provide a comprehensive assessment of its accuracy.\n",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Final Evaluation",
      "content": "Final evaluation of submissions will be based on the F1 score to measure the precision and recall in predicting the direction of cryptocurrency prices, either up or down. Due to the public availability of all data sets and to prevent lookahead bias or falsification of results, it is mandatory for participants to submit their code via an open repository on GitHub or GitLab. This requirement enables us to independently run and verify each model’s training and evaluation processes on additional data. Entries that do not comply with this open repository submission guideline will be ineligible for consideration in the competition.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}