Prediction of Factory Electric Consumption

This repository contains a regression solution for the Kaggle competition on Factory Electric Consumption. The goal is to accurately predict the electric power consumption of a factory using historical sensor and usage data.

Project Objective

To develop a regression model capable of predicting future electric consumption in a factory environment based on various numerical and categorical features. Accurate predictions can assist in optimizing energy usage, reducing operational costs, and enhancing sustainability.

Dataset Summary

• Training Set: 13 872 records
• Test Set: 2 160 records (target to predict)
• Target Variable: Electric_Consumption
• Evaluation Metric: RMSE (Root Mean Squared Error)

Models Used

• Linear Regression
• Polynomial Regression (with hyperparameter tuning)
• Random Forest Regressor (with hyperparameter tuning)
• XGBoost Regressor (with hyperparameter tuning)

Project Structure

• Libraries Used
• Data Visualization and Exploration
• Data Processing:
  • Feature Extraction
  • Outliers Analysis
  • Checks for missing values
  • Encoding of non-numerical features
• Models Training and Evalutaion: Different models were tried: Linear Regression, Polynomial Regression, Random Forest Regressor, XGBoost Regressor. For each model it was performed:
  • Cross-Validation RMSE
  • Hyperparameters tuning using GridSearchCV
  • Feature Importance analysis
  • RMSE on the validation set
  • Post-processing to ensure predictions are non-negative
• Results and Comparison

Model Evaluation

Models were evaluated using cross-validation (CV) and validation set performance:

Model	CV RMSE	Std. Dev.	Validation RMSE	Best Parameters
Linear Regression	3.476	0.199	3.192	Default
Polynomial Regression	3.064	1.112	2.241	{'polynomialfeatures__degree': 2, 'polynomialfeatures__include_bias': True}
Random Forest	2.434	0.450	1.674	{'max_depth': None, 'min_samples_split': 2, 'n_estimators': 200}
XGBoost	2.520	0.430	1.494	{'learning_rate': 0.05, 'max_depth': 6, 'n_estimators': 1000, 'predictor': 'cpu_predictor', 'subsample': 0.8, 'tree_method': 'hist'}

References

• Kaggle Competition

Contents

• Software_and_Data/ – Contains the complete solution:
  • Python Notebook (.ipynb and .pdf)
  • Setup instructions and environment requirements
  • Data/ – Includes training, test data and a submission folder with the different predictions
  • README.md – Project description
• Doc/ - Contains:
  • Slides for the presentation (.tex and .pdf)
• README.md – Project overview