Store Sales - Time Series Forecasting

📌 Project Overview

Retail sales forecasting is essential for optimizing inventory, managing supply chains, and increasing revenue. This project predicts sales for thousands of product families sold in Favorita stores in Ecuador. Using machine learning techniques like SGDRegressor, LSTM (Long Short-Term Memory Networks), and SVM (Support Vector Machine), we aim to improve sales prediction accuracy.

🏗 Dataset Description

The dataset consists of multiple CSV files:

• train.csv → Historical sales data for different product families across stores.
• test.csv → Contains the same structure as train.csv but without sales values (used for prediction).
• stores.csv → Metadata about store locations and types.
• oil.csv → Daily oil price data (Ecuador’s economy depends on oil prices).
• holidays_events.csv → List of holidays and events that may impact sales.
• transactions.csv → Number of daily transactions for each store.

🛠 Technologies & Libraries Used

• Python (Data Analysis & Modeling)
• Jupyter Notebook (Code Execution)
• Pandas, NumPy (Data Manipulation)
• Matplotlib, Seaborn (Data Visualization)
• Scikit-Learn (Machine Learning - SGDRegressor, SVM)
• TensorFlow, Keras (Deep Learning - LSTM)
• TQDM (Progress tracking)

🚀 How to Run the Project

1️⃣ Install Dependencies

Run the following command in your terminal:

pip install pandas numpy matplotlib seaborn scikit-learn tensorflow tqdm

2️⃣ Clone the Repository & Navigate to the Folder

git clone https://github.com/yourusername/store-sales-time-series-forecasting.git
cd store-sales-time-series-forecasting

3️⃣ Run the Jupyter Notebook

jupyter notebook store_sales_forecasting.ipynb

🔍 Exploratory Data Analysis (EDA)

We performed the following analyses to understand the dataset:

• Sales Trends Over Time 📈
• Impact of Promotions on Sales 🛒
• Effect of Holidays on Revenue 🎉
• Oil Price vs. Sales Correlation ⛽

🧠 Machine Learning Models

We implemented three models to predict sales:

1️⃣ SGDRegressor (Fast ML Model)

• Applied StandardScaler to normalize features.
• Trained a Stochastic Gradient Descent (SGD) Regressor.
• Performance Metric: Mean Squared Error (MSE)

2️⃣ LSTM (Deep Learning Model for Time-Series)

• Preprocessed data using MinMaxScaler.
• Reshaped input to (samples, time-steps, features) format.
• Built an LSTM network with 2 layers.
• Trained using Adam optimizer.

3️⃣ SVM (Support Vector Machine)

• Standardized features using StandardScaler.
• Implemented SVR (Support Vector Regression) with RBF kernel.
• Performance Metric: Mean Squared Error (MSE).
• Note: SVM was not yet executed in the notebook.

📊 Results & Comparison

Model	Mean Squared Error (MSE)
SGDRegressor	808,277.18
LSTM	11,396.67
SVM	(Pending Execution)

🔹 LSTM performed significantly better than SGDRegressor because it captures time-series dependencies more effectively.

📂 Project Structure

Store_Sales_Time_Series_Forecasting/
│
├── data/
│   ├── train.csv
│   ├── test.csv
│   ├── stores.csv
│   ├── oil.csv
│   ├── holidays_events.csv
│   ├── transactions.csv
│
├── notebooks/
│   ├── store_sales_forecasting.ipynb
│
├── models/
│   ├── sgd_model.pkl
│   ├── lstm_model.h5
│   ├── svm_model.pkl (Pending Execution)
│
├── images/
│   ├── sales_trend.png
│   ├── model_comparison.png
│
├── README.md
└── requirements.txt

🤝 Contributing & Future Work

Future Improvements:

✅ Try XGBoost or Random Forest models for better accuracy. ✅ Hyperparameter tuning for LSTM to further optimize predictions. ✅ Execute and analyze the performance of SVM. ✅ Deploy the model as a web app using Flask or FastAPI.

Want to Contribute?

Feel free to fork the repository, make improvements, and submit a pull request!

📧 Contact:_____ | 💻 GitHub: BabaMalik

---

🚀 Thank you for checking out the project! Hope it helps in understanding store sales forecasting! 🔥

https://www.kaggle.com/competitions/store-sales-time-series-forecasting/data