This repository contains a project that predicts daily sales for Rossmann stores using a machine learning model. It provides a REST API for real-time predictions and includes scripts for data preprocessing, model training, evaluation, and deployment.

Rossmann is a chain of drugstores, and this project aims to predict daily sales across its various stores. Accurate sales predictions enable better decision-making and resource allocation, including inventory management, staffing, and promotional activities. The project follows these main tasks:

1. Data Cleaning and Preprocessing
2. Exploratory Data Analysis (EDA)
3. Feature Engineering and Model Building
4. Model Evaluation and Interpretation
5. Model Serialization and Serving with FastAPI
6. Deployment for Real-Time Predictions

• Handle missing values and outliers.
• Create new features, such as weekday, weekends, and holiday proximity.
• Scale numerical features and encode categorical variables.

• Analyze sales trends and correlations.
• Visualize customer purchasing behavior.
• Investigate the impact of promotions and competitors.

• Implement a pipeline using scikit-learn for preprocessing and modeling.
• Train a Random Forest Regressor model to predict daily sales.

• Build a Long Short-Term Memory (LSTM) model for time series sales prediction.
• Use TensorFlow/Keras for deep learning modeling.

• Serve the trained models using a REST API built with FastAPI.
• Provide endpoints for making predictions based on store features.

• Deploy the API to a cloud platform for production use.
• Use tools like DVC for model tracking and Git for version control.

project/
├── data/                       # Contains raw and cleaned datasets
│   ├── train.csv
│   ├── test.csv
│   ├── store.csv
├── models/                     # Serialized models
│   └── sales_model.pkl
├── notebooks/                  # Jupyter notebooks for EDA and experiments
│   ├── data_cleaning.ipynb
│   ├── eda.ipynb
│   ├── modeling.ipynb
├── scripts/                    # Python scripts for tasks
│   ├── cleaning.py             # Data cleaning and preprocessing
│   ├── eda.py                  # Exploratory data analysis
│   ├── modeling.py             # Model training and evaluation
│   ├── deep_learning.py        # LSTM modeling
│   ├── api.py                  # FastAPI implementation
├── requirements.txt            # Dependencies
├── README.md                   # Project documentation
├── Procfile                    # Deployment configuration for Heroku
└── dvc.yaml                    # DVC pipeline configuration

• Python 3.8+
• Virtual environment (optional but recommended)

git clone https://github.com/Hunegn/rossmann-sales-prediction.git
cd rossmann-sales-prediction

pip install -r requirements.txt

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

Place the datasets (train.csv, test.csv, store.csv) in the data/ directory.

uvicorn scripts.api:app --reload

Visit http://127.0.0.1:8000 for the API health check.

Send a POST request to /predict/ with the following JSON body:

{
  "CompetitionDistance": 500.0,
  "Promo": 1,
  "Weekday": 3,
  "DaysToHoliday": 10,
  "StoreType": "a",
  "Assortment": "basic",
  "MonthPhase": "Beginning"
}

Response:

{
  "prediction": 12345.67
}

• Optimize model hyperparameters for better performance.
• Explore other algorithms like XGBoost or LightGBM.
• Improve deep learning model with more advanced architectures.
• Add authentication to the API for secure access.

Contributions are welcome! Please fork the repository and submit a pull request with your changes.

This project is licensed under the MIT License.