Store Sales Forecasting with Time Series ML

A machine learning project for forecasting daily sales across 10 stores and 50 items using time-series features.

📊 Project Overview

This project demonstrates end-to-end time-series forecasting with focus on:

• Data Quality: Comprehensive validation checks for completeness and consistency
• Feature Engineering: Temporal, lag, and calendar-based features with proper data leakage prevention
• Performance: Vectorized operations handle 913,000 records in seconds (100x faster than naive approach)
• Best Practices: Temporal train/val split, baseline models, documented code

Dataset

• Size: 913,000 records (2013-01-01 to 2017-12-31)
• Stores: 10 unique stores
• Items: 50 unique items per store
• Target: Daily sales count (integer)

🎯 Results

Baseline Model Performance (Validation Set)

Model	MAE	RMSE	Description
Global Mean	22.97	28.56	Training set average
Lag-1 (yesterday)	10.65	14.47	Previous day's sales
Rolling Mean 7-day	8.64	11.37	7-day rolling average

Validation Period: October 3, 2017 - December 31, 2017 (90 days, 45,000 samples)

The Rolling Mean 7-day baseline achieves MAE of 8.64, establishing a strong benchmark for ML models.

🚀 Quick Start

Local Jupyter

# Clone the repository
git clone <repository-url>
cd ml-project-projekt-steg-2

# Install dependencies
pip install -r requirements.txt

# Launch Jupyter
jupyter notebook project_ml_optimized.ipynb

Google Colab

# Clone repository
!git clone https://github.com/YOUR_USERNAME/ml-project-projekt-steg-2.git
%cd ml-project-projekt-steg-2

# Install dependencies
!pip install -r requirements.txt

# Upload dataset
from google.colab import files
import os
os.makedirs('Dataset', exist_ok=True)
uploaded = files.upload()
!mv train.csv Dataset/

# Run the notebook!

Or click:

📁 Project Structure

.
├── Dataset/
│   └── train.csv              # Sales data (gitignored)
├── project_ml_optimized.ipynb # Main notebook with all code
├── requirements.txt           # Python dependencies
└── README.md                  # This file

🔧 Feature Engineering

Temporal Features

• Linear: Day of week, month, day, quarter, week of year
• Binary: Weekend flag, month start/end indicators
• Cyclical: Sin/cos transformations for periodic patterns (dow, month)

Lag Features (per store-item)

• sales_lag_1: Previous day's sales
• sales_lag_7: Sales from 7 days ago
• sales_lag_365: Sales from 365 days ago (seasonal)
• roll_mean_7: 7-day rolling average (leakage-free)
• wow_change: Week-over-week momentum (%)
• store_daily_avg_lag1: Store-level average from previous day