README.md

metadata

title: Sales Forecaster - Kaggle Sticker Challenge
emoji: 📊
colorFrom: blue
colorTo: green
sdk: streamlit
sdk_version: 1.32.0
app_file: app.py
pinned: false
python_version: '3.10'
license: gpl-3.0

Sticker Sales Prediction Project

📌 Project Overview

This project predicts the sales volume (num_sold) for various Kaggle-branded products across different stores and countries. It was developed as part of a career transition into Data Science, focusing on Time-Series regression and feature engineering.

🚀 Technical Highlights

• Model: XGBoost Regressor.
• Metric: Achieved a 6.42% MAPE (Mean Absolute Percentage Error) on the validation set.
• Key Strategy: Utilized Log Transformation ($y = \log(1+x)$) to handle the right-skewed distribution of sales data, which improved the MAPE from 16% to 6.4%.
• Feature Engineering: Extracted temporal features such as day_of_week, is_weekend, and month to capture seasonality and the "Weekend Effect."

📂 File Structure

• app.py: Streamlit application code.
• xgboost_model.json: Pre-trained XGBoost model.
• model_columns.pkl: List of features used during training (to ensure input alignment).
• requirements.txt: Python dependencies.

🛠️ Local Setup

To run this project locally with Python 3.10:

1. Clone the repository.
2. Create a virtual environment:

   python -m venv venv
   source venv/bin/activate  # On Windows: venv\Scripts\activate