Predicting sales accurately at the product level is challenging due to uncertainty caused by various factors, including seasonality, marketing campaigns, holidays, or other contextual information. Another challenge of modeling retail data is making decisions based on limited history. This project aims to overcome these challenges by building a reliable sales forecasting model.
- Develop an Accurate Sales Prediction Model for Walmart: The main aim is to create a precise sales prediction model tailored for Walmart, utilizing historical sales data and advanced algorithms.
- Account for Holiday Markdown Effects: One of the key challenges in this project is dealing with a significant portion of missing data in the markdown columns. My goal is to develop strategies and techniques to effectively handle this missing data and ensure the accuracy and reliability of our sales prediction model despite this limitation.
- Analyze Sales Patterns Over Time: Conduct a comprehensive analysis of long-term sales trends, seasonality, and fluctuations to provide valuable insights and adapt the prediction model to changing market conditions.
The dataset for this project is sourced from Kaggle and contains historical weekly sales data from 45 Walmart stores, spanning 81 different departments across various regions from 2010-02-05 to 2012-11-01.
Download the dataset here
- Jupyter Notebook: Data Wrangling, Exploratory Data Analysis, Preprocessing and Modeling, SHAP Analysis
- Project Final Report and Presentation