Sales Forecasting using Machine Learning

Project Overview

This project focuses on forecasting future sales using historical retail data from the Superstore dataset. A Random Forest regression model is used along with time-series feature engineering to capture patterns and generate future predictions.

Key Features

• Time series forecasting on real-world retail data

• Feature engineering:

  • Extraction of month and year
  • Lag features (lag1, lag2)
  • Rolling mean (3-step average)

• Random Forest Regressor

• Train-test split without shuffling (time-series aware)

• Confidence interval visualization

• Multi-step forecasting for the next 12 months

Visual Outputs

The following visualizations are generated and stored in the outputs/ directory:

• Sales trend over time
• Actual vs predicted comparison
• Forecast with confidence interval
• Future sales forecast

Tech Stack

• Python
• Pandas
• NumPy
• Matplotlib
• Scikit-learn

Project Structure

FUTURE_ML_01/
│
├── data/
│   └── Sample - Superstore.csv
├── outputs/
│   ├── advanced_forecast.png
│   ├── data_plot.png
│   └── future_forecast.png
├── src/
│   ├── Sample - Superstore.csv
│   └── task1.py
├── .gitignore
├── LICENSE
├── README.md

How to Run

Run the main script from the project root directory:

python src/task1.py

Results

The model is able to capture underlying trends in the sales data and generate reasonable forecasts. The use of lag features and rolling statistics improves predictive performance on time-dependent data.

Author

Samarpit Gujral