# Forecasting Retail Purchases — Data

来源：https://www.kaggle.com/competitions/forecasting-retail-purchases/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

The **"Forecasting Retail Purchases"** competition aims to predict whether customers will make purchases during a specified time period based on their historical purchasing behavior. The prediction window is set from **April 7, 2024, to July 30, 2024**, and participants are provided with customer purchase data from **January 2, 2023, to April 6, 2024** to build their models.

### Dataset Files

The competition consists of four main CSV files:

1. **train.csv**:
   - This file contains the training data used to train predictive models. 
   - **Columns**:
     - **user**: An integer identifier for each customer.
     - **time**: The date associated with each observation in the format YYYY-MM-DD.
     - **bought**: A binary indicator where `1` represents that the customer made a purchase on that date, and `0` signifies no purchase.
   - The training data spans all customer activity before April 7, 2024.

2. **test.csv**:
   - This file holds the test data on which participants will make their predictions.
   - **Columns**:
     - **user**: An integer identifier for each customer.
     - **time**: The date associated with each observation in the format YYYY-MM-DD.
   - The test data corresponds to the period from April 7, 2024, to July 30, 2024. Note that this file does not include the **bought** column, as the goal is to predict this value.

3. **sample_submission.csv**:
   - This file serves as a template for participants to format their predictions.
   - **Columns**:
     - **ID**: A unique identifier combining `user` and `time` (in the format **user_time**).
     - **Usage**: Randomly assigned values of either "Public" or "Private," indicating the sample of public and private test sets.
     - **bought**: The actual purchase outcome (`1` or `0`) for the respective user on that date.

### Objective

Participants are tasked with predicting whether each customer will make a purchase on specific dates within the testing period, utilizing the historical purchase data provided in the training set. The models will be evaluated based on their accuracy in predicting the `bought` variable, enabling participants to refine their forecasting skills in a real-world retail context.