# Global E-Commerce Customer Purchase Prediction — Data

来源：https://www.kaggle.com/competitions/global-e-commerce-customer-purchase-prediction/data

采集时间：2026-10-19T09:42:26Z

# 📂 Dataset Description

This competition provides a large-scale, multi-table dataset simulating a global e-commerce platform. The data reflects realistic production challenges including messy formatting, inconsistent categorical values, missing data, high-cardinality features, and distribution shifts.

Participants must integrate and engineer features across multiple relational tables to build customer-level predictive models.

---

# 🎯 What Are You Predicting?

For each `customer_id` in the test set, you must predict:

> The probability that the customer will make a purchase within the next 30 days.

This is a **binary classification problem**, and predictions must be probabilities between 0 and 1.

---

# 📁 Files

## Core Competition Files

- **train.csv** — Training labels  
- **test.csv** — Test set for prediction  
- **sample_submission.csv** — Example submission file  

## Relational Data Tables

- **customers.csv** — Customer profile and demographic attributes  
- **sessions.csv** — Session-level browsing logs  
- **transactions.csv** — Historical purchase records  
- **marketing_campaigns.csv** — Campaign metadata  
- **geo_data.csv** — Regional socioeconomic attributes  

Participants must join and aggregate these tables to create modeling features.

---

# 📄 File Details

## 1️⃣ train.csv

Contains customer IDs and training labels.

### Columns:

- `customer_id` — Unique customer identifier  
- `target` — Binary label  
  - 1 = Customer purchased within next 30 days  
  - 0 = No purchase  

---

## 2️⃣ test.csv

Contains customer IDs for which predictions must be generated.

### Columns:

- `customer_id` — Unique customer identifier  

The target column is not included.

---

## 3️⃣ sample_submission.csv

Provides the correct submission format.

### Format:
customer_id,target 100001,0.8234 100002,0.1452
---

## 4️⃣ customers.csv

Customer-level profile and demographic information.

### Key Columns:

- `customer_id` — Unique ID  
- `age` — May contain numeric values or non-numeric entries  
- `country_code` — Inconsistent formatting possible  
- `region` — Geographic region identifier  
- `signup_date` — Customer registration date  
- `loyalty_score` — Stored as text  
- `email_open_rate` — Email engagement metric  
- `discount_usage_rate` — Discount behavior metric  
- `avg_review_score` — Historical review average  
- `referral_code` — High-cardinality identifier  
- `customer_tier` — Loyalty program tier  
- `credit_limit` — Currency-formatted string  
- Additional noise columns  

---

## 5️⃣ sessions.csv

Each row represents one browsing session.

### Key Columns:

- `session_id` — Unique session identifier  
- `customer_id` — Foreign key  
- `session_timestamp` — Timestamp of session  
- `session_duration` — Session duration  
- `pages_viewed` — May contain non-numeric values  
- `cart_additions` — Number of cart additions  
- `bounce_flag` — Mixed-format indicator  
- `traffic_source` — Acquisition channel  
- `device_type` — Device category  
- `campaign_id` — Associated marketing campaign  
- `geo_ip_region` — Geographic region  

This table contains millions of rows and requires aggregation.

---

## 6️⃣ transactions.csv

Historical purchase records.

### Key Columns:

- `transaction_id` — Unique transaction identifier  
- `customer_id` — Foreign key  
- `transaction_timestamp` — Purchase time  
- `order_value` — Currency-formatted string  
- `items_count` — Number of items purchased  
- `payment_method` — Payment type  
- `discount_applied` — Mixed-format flag  
- `shipping_speed` — Delivery type  
- Additional noise columns  

---

## 7️⃣ marketing_campaigns.csv

Marketing campaign metadata.

### Columns:

- `campaign_id` — Unique campaign identifier  
- `campaign_type` — Campaign category  
- `campaign_budget` — Currency-formatted value  
- `region_target` — Target region  
- `start_date`  
- `end_date`  

---

## 8️⃣ geo_data.csv

Regional socioeconomic indicators.

### Columns:

- `geo_ip_region` — Region key  
- `average_income`  
- `urban_ratio`  
- `internet_penetration`  
- `region_tier`  

---

# 📊 Data Characteristics

Participants should expect:

- Missing values  
- Numeric fields stored as strings  
- Currency symbols and formatted numbers  
- Inconsistent categorical casing  
- High-cardinality categorical variables  
- Potential distribution differences between train and test  
- Noise features  

Not all columns are predictive.

---

# 📚 Common Acronyms

- **ROC-AUC** — Receiver Operating Characteristic Area Under Curve  
- **CV** — Cross-Validation  
- **ID** — Identifier  
- **LLM** — Large Language Model (allowed under rules)  

---

# ⚠ Important Notes

- This is a relational dataset.  
- Aggregation and feature engineering are required.  
- Avoid data leakage when merging transaction and session information.  
- Robust validation strategies are strongly recommended.  

Participants are responsible for careful preprocessing, feature engineering, and modeling.

Good luck!