📦 eCommerce behavior data from multi category store

An advanced analysis of eCommerce customer behavior using machine learning and deep learning models on real-world data from a large multi-category online store.

Dataset

Source: Kaggle Dataset

• Size: ~285 million events over 7 months (Oct 2019 - Apr 2020)
• Event Types:
  • view - Product viewed
  • cart - Added to cart
  • remove_from_cart - Removed from cart
  • purchase - Purchased product
• Features:
  • event_time, event_type, product_id, category_id, category_code
  • brand, price, user_id, user_session

🎯 Objective

Predict customer behavior (view, add to cart, purchase) during online shopping sessions using time series forecasting, machine learning, and deep learning.

---

🛠️ Workflow

1️⃣ Data Preprocessing

• Missing values imputation
• Linear interpolation for gaps
• Detrending & deseasonalization
• Train/Test Split (80/20)

2️⃣ Visualization

• Price distribution (Boxplots, Histograms, KDE)
• Event frequency over time
• Seasonal decomposition (Additive & Multiplicative)
• ACF & PACF plots for time dependencies

---

🤖 Models Used

Machine Learning Models

Model	Accuracy
Random Forest	50%
Logistic Regression	43%
CatBoost	50%
XGBoost	50%

Deep Learning Models

Model	Accuracy
LSTM	96.05%
GRU	96.06%
Transformer	96.05%
Temporal CNN (TCN)	96.05%

---

⚙️ Optimization

• ML Models: Grid Search, Random Search
• DL Models: Adam Optimizer, Dropout, Label Encoding
• Evaluation with Confusion Matrix, Accuracy, Precision, Recall, F1-Score

---

📊 Results

• ✅ Best Machine Learning Model: XGBoost (50%)
• ✅ Best Deep Learning Model: GRU (96.06%)
• Deep learning models significantly outperformed machine learning models in predictive accuracy.

---

📝 Conclusion

Combining time series forecasting, machine learning, and deep learning techniques provides a robust predictive system for understanding customer behavior in eCommerce. GRU showed the best balance between accuracy and efficiency.

---

📚 References

• Dataset: Open CDP Project
• Special thanks to REES46 Marketing Platform for providing the dataset.