📈 Retail SKU Demand Prediction

📌 Overview

This project implements an end-to-end demand forecasting system to predict SKU-level daily demand for a retail store over a 21-day future horizon.
The solution uses classical machine learning / statistical techniques, applies standard time-series feature engineering, and evaluates performance using RMSE.

---

📂 Dataset

The dataset contains transaction-level retail sales data for the financial years 2009–10 and 2010–11.

Column Description

• InvoiceNo: 6-digit invoice number (starting with C indicates cancellation)
• StockCode: 5-digit SKU identifier
• Description: Product name
• Quantity: Quantity purchased per transaction
• InvoiceDate: Date and time of transaction
• UnitPrice: Price per unit (GBP)
• CustomerID: 5-digit customer identifier
• Country: Customer’s country

---

🎯 Objective

• Forecast daily demand at SKU level
• Predict demand for the next 21 days
• Apply time-series feature engineering
• Split data into train / validation / test
• Evaluate accuracy using RMSE
• Export forecasts as a SKU × Day matrix (CSV)

---

🧠 Methodology

1️⃣ Data Cleaning & Preprocessing

• Converted InvoiceDate to datetime format
• Removed cancelled transactions (InvoiceNo starting with C)
• Removed invalid records (negative quantity or price)
• Filtered valid 5-digit SKUs
• Aggregated transaction data to daily SKU-level demand
• Filled missing dates with zero demand to ensure time continuity
• Removed weak SKUs with very low overall demand

---

2️⃣ Feature Engineering

• Lag features (e.g., 1, 7, 14 days)
• Rolling statistics to capture short-term trends
• Time-based features to capture seasonality

---

3️⃣ Model Training & Forecasting

• Used a classical ML / statistical forecasting model
• Performed time-aware splitting into:
  • Training set
  • Validation set
  • Test set
• Forecasted demand iteratively for a 21-day horizon

---

4️⃣ Evaluation

• Evaluated model performance using Root Mean Squared Error (RMSE)
• RMSE calculated on the test dataset
• Results analyzed at SKU level

---

5️⃣ Output

• Forecasts generated for 21 future days
• Output stored as a CSV file in the following format:

SKU	Day_1	Day_2	...	Day_21
12345	10	12	...	15

Each value represents the forecasted quantity for the SKU on that day.

---

🛠 Tools & Technologies

• Python
• Pandas, NumPy
• Scikit-learn / classical ML models
• Matplotlib / Seaborn
• Jupyter Notebook