{
  "fetched_at": "2026-10-19T07:20:23Z",
  "source_url": "https://www.kaggle.com/competitions/anomaly-class-detection/data",
  "pages": [
    {
      "name": "data-description",
      "content": "## Imbalance of the dataset\n\nThe given dataset for this competition is characterized by a significant class imbalance. This means that the distribution of the target variable, which represents the movement of stock prices, is highly skewed towards one or a few classes, while other classes have relatively fewer instances. It is important to be aware of this imbalance when developing your models and evaluating their performance. Consider applying appropriate techniques such as class weighting, oversampling, or undersampling to address this challenge and ensure that your models are capable of effectively capturing patterns across all classes.\n\n## Columns\n\n*   `result` - defines this way:\n - 0 - short/long/anomaly;\n - 1 - hold;\n\n#### Categorical Columns\n\n*   `RSI_signal`\n*   `Stoch_O_signal`\n*   `ichimoku_c_signal`\n*   `fibonacci_signal`\n*   `Bollinger_signal`\n*   `exchange`\n\n#### Numeric Columns\n\nBasically they are all other columns in dataset. But I need to say a few special words about columns below:\n\n*   `ichimoku_c_conversion_line`\n*   `ichimoku_c_base_line`\n*   `ichimoku_c_leading_span_a`\n*   `ichimoku_c_leading_span_b`\n*   `pivot_support_1`\n*   `pivot_resistance_1`\n*   `pivot_support_2`\n*   `pivot_resistance_2`\n*   `pivot_support_3`\n*   `pivot_resistance_3`\n\nThe mentioned indicators represent various points of interest in price movement analysis. These indicators help traders and analysts identify potential areas of price reversals, support, and resistance, enabling them to make more informed decisions in their trading strategies.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "Welcome to the stock market anomaly detection competition! \n\nIn this challenge, you'll be working with a dataset that reflects the unique characteristics of the stock market. **Here, the majority of deals (99%) are classified as \"hold,\" while the remaining 1% comprises both short and long positions**. However, for the purpose of this competition, the distinction between short and long is not significant. What truly matters is **identifying transactions that are not classified as \"hold\".** These non-hold transactions are represented as a single class labeled as \"1\"\n\n# Problem Statement:\nDetecting anomalies in the stock market is a critical task with various real-world applications. Given the **highly imbalanced** nature of the dataset, where the \"hold\" class significantly outweighs the \"1\" class, accurately identifying these non-hold transactions becomes a challenging yet essential objective. This competition provides a platform for you to demonstrate your skills in developing robust anomaly detection models specifically tailored for the stock market.\n\n# Dataset:\nThe dataset provided is carefully curated, encompassing a range of features and capturing the dynamics of the stock market. With the majority of transactions falling under the \"hold\" class, the dataset's imbalance poses a significant hurdle in accurately detecting the non-hold anomalies. Exploring and analyzing the dataset thoroughly will be crucial in gaining insights and devising effective strategies for anomaly detection.\n\n# Objective:\nYour ultimate goal in this competition is to build innovative models that can accurately identify non-hold anomalies within the stock market dataset. Submissions will be evaluated based on key anomaly detection metrics that account for the class imbalance, such as precision, recall, and **F1-score**.\n\n# Evaluation Metrics:\n**F1-score** serves as the evaluation metrics for this competition. This metric provides a comprehensive assessment of your model's performance, considering the challenges posed by the dataset's class imbalance. Achieving a balance between high precision (accurately detecting true non-hold anomalies) and high recall (capturing as many non-hold anomalies as possible) will be instrumental in developing strong anomaly detection models.\n\n# Get started:\nTo excel in this competition, you are encouraged to explore various anomaly detection techniques and machine learning algorithms. Consider leveraging unsupervised learning methods, such as clustering or outlier detection algorithms, as well as supervised approaches customized for imbalanced datasets. Additionally, feature engineering and data visualization techniques may prove valuable in extracting meaningful patterns from the stock market dataset.\n\n**Best of luck, and may your models successfully uncover the non-hold anomalies hidden within the stock market dataset!**",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "The evaluation metric for this competition is the F1-score with macro averaging. This metric provides a balanced assessment of a model's performance across all classes, taking into account both precision and recall. Participants should aim to optimize their models for high macro F1-scores to achieve strong overall performance.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "Anything you want!",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}