Revenue Forecast - Datathon 2026

This project utilizes a LightGBM machine learning model combined with hyperparameter optimization via Optuna to forecast revenue based on historical data.

📁 Project Structure

Ensure your project structure looks like this before running:

Datathon/
│
├── data/                       # Contains raw data (you need to place the csv files here)
│   ├── sales.csv               # Training data
│   ├── sales_test.csv          # Test data to predict
│   └── sample_submission.csv   # Sample format for submission
│
├── src/                        # Contains Python source code
│   ├── Feature_engineering.py  # Script for data preprocessing and feature extraction
│   └── Main.py                 # Script for training the LightGBM model and generating predictions
│
├── out/                        # Auto-generated directory for predictions and plots
│
├── requirements.txt            # List of required Python libraries
└── run.bat                     # Batch script to automatically run the entire pipeline on Windows

🚀 How to Run

Method 1: Automated Run (Recommended for Windows)

Simply double-click the run.bat file in the root directory. This script will automatically:

1. Install all necessary dependencies from requirements.txt.
2. Navigate into the src folder.
3. Execute Feature_engineering.py to process the raw data and generate feature datasets.
4. Execute Main.py to search for optimal hyperparameters using Optuna, train the LightGBM model, and output the final results.

Method 2: Manual Run via Terminal

Open a Terminal (Command Prompt / PowerShell / bash) at the project root directory (Datathon/):

1. Install dependencies:

   pip install -r requirements.txt

2. Execute Python scripts (Note: you must run these from inside the src directory):

   cd src
   python Feature_engineering.py
   python Main.py

📊 Outputs

Upon successful execution, an out/ directory will be created (if it doesn't already exist) containing the following files:

• submission.csv: The final prediction file containing the revenue forecast for submission.
• LightGBM_Simple_Revenue_Forecast.png: A plot comparing the actual revenue (Training set) with the predicted revenue (Test set).
• Feature_Importance.png: A bar chart illustrating the Feature Importance of the top 30 input variables learned by the model.