README.md

---
license: cc-by-4.0
language:
- en
- ru
tags:
- time-series
- finance
- bank
pretty_name: Alfa Card Transactions
---

### Dataset Summary

Alfa Card Transactions is a unique high-quality dataset collected from real data sources of Alfa Bank's clients' transactions for the task of the default prediction. It consists of histories of transactions, IDs of credit products and flags of corresponfing default. 

### Supported Tasks and Leaderboards

The dataset is supposed to be used for training models for the classical bank task of predicting the default of the applicant. 

## Dataset Structure

### Data Instances

The example of one sample is provided below
```
{
  'app_id': 10,
  'transactions':
  [ 
    [10.0, 0.0, 1.0, 6.0, 54.0, 22.0, 3.0, 1.0, 2.0, 2.0, 2.0, 1.0, 66.0, 2.0, 2.0, 0.0, 351.0, 50.0,-1.0, 1.0],
    [10.0, 0.3876771200456198, 1.0, 2.0, 54.0, 8.0, 1.0, 1.0, 2.0, 1.0, 2.0, 1.0, 66.0, 2.0, 2.0, 21.0, 351.0, 50.0, 21.0, 2.0]
  ],
  'product': 1,
  'flag': 0
}
```

### Data Fields

- `app_id`: application ID.
- `history`: an array of transactions where each transaction is represented as a 20-dimensional array, each element of the array represents a corresponding feature from the following list.
  - `app_id`: application ID.
  - `amnt`: normalized transaction amount. 0.0 - corresponds to omissions.
  - `currency`: transaction currency ID.
  - `operation_kind`: ID of the transaction type.
  - `card_type`: unique identifier of the card type.
  - `operation_type`: ID of the type of plastic card transaction.
  - `operation_type_group`: ID of a group of card transactions, for example, debit card or credit card.
  - `ecommerce_flag`: feature of e-commerce.
  - `payment_system`: ID of the payment system type.
  - `income_flag`: feature of debiting/depositing funds to the card.
  - `mcc`: unique identifier of the type of outlet.
  - `country`: transaction country ID.
  - `city`: transaction city ID.
  - `mcc_category`: ID of the transaction store category.
  - `day_of_week`: day of the week when the transaction was made.
  - `hour`: hour when the transaction was made.
  - `days_before`: number of days before the date of issue of the loan.
  - `weekofyear`: number of the week in the year when the transaction was made.
  - `hour_diff`: number of hours since the last transaction for this client.
  - `transaction_number`: sequence number of the client's transaction.
- `product`: product ID for which it is necessary to make a decision whether the applicant will go into default or not
- `flag`: target, 1 - the fact of going into default.