🛍️ Retail Sales Prediction with Random Forest

This project builds and evaluates various Random Forest models to predict retail sales based on time, holiday, and store-specific features. It includes preprocessing, model training with hyperparameter tuning, evaluation, and prediction on new datasets.

📂 Files

• csv_files/sales.csv — Main dataset used for model training.
• csv_files/ironkaggle_notarget.csv — New data for predictions (without target).
• csv_files/ironkaggle_solutions.csv — Comparison/solution dataset.
• models/ — Folder containing saved models (.pkl format).

🧠 Machine Learning Models

Three main Random Forest models were trained:

✅ model_rf01

• Features: All features except sales
• Normalization: StandardScaler
• Hyperparameters: n_estimators=100, max_depth=20
• Status: Trained and saved to models/model_rf01.pkl

✅ model_rf02

• Features: Dropped open, is_weekend, school_holiday, year
• Normalization: StandardScaler
• Hyperparameters: n_estimators=100, max_depth=None
• Status: Trained and saved to models/model_rf02.pkl

✅ model_rf03

• Grid Search (classification mistake — used scoring='f1')
• Status: Trained, but improperly scored for regression

⚙️ HalvingGridSearchCV (Experimental)

• Feature engineering and time estimate logic for efficient hyperparameter search
• Output model: models/best_rf_halving.pkl
• Status: Time estimation and progress tracking using tqdm & threading

🤖 Other Models Tried

Several other machine learning models were imported (e.g., XBoost, GradientBoostingRegressor, lightgbm etc.) and briefly tested. However, they performed significantly worse than Random Forest on this dataset.
As a result, we removed their implementations to keep the project clean and focused on the best-performing model.

📊 Evaluation Metrics

Each model is evaluated using:

• MAE: Mean Absolute Error
• RMSE: Root Mean Squared Error
• R² Score: Coefficient of Determination

🔍 Feature Importance

Feature importances were plotted using model.feature_importances_ to understand which variables most influence predictions.

📈 Correlation Matrix

A masked heatmap using seaborn highlights correlations between features.

📤 Prediction on New Data

The file ironkaggle_notarget.csv was:

• Preprocessed with the same steps as training data
• Normalized using the previously fitted StandardScaler
• Predicted using model_rf02
• Output stored in df01_with_predictions.csv

🛠️ Tools Used

• pandas, numpy, matplotlib, seaborn
• scikit-learn: RandomForestRegressor, GridSearchCV, HalvingGridSearchCV
• joblib: For model saving/loading
• tqdm and threading: For progress monitoring

📝 How to Run

1. Ensure required libraries are installed (requirements.txt or use `pip install -r requ

👥 Contributors

We welcome contributions from anyone interested in improving this project! Here are the current contributors:

• Author: JorgeMMLRodrigues

• Email: jorgemmlrodrigues@gmail.com

• GitHub: https://github.com/JorgeMMLRodrigues

• Author: Mic-dev-gif

• Email: michelemontalvo@outlook.com

• GitHub: https://github.com/Mic-dev-gif

• Email: simiatawane@gmail.com

• Email: felipe.rocha@ironhack.com