International Research Journal of Engineering and Technology (IRJET)
e-ISSN: 2395-0056
Volume: 12 Issue: 03 | Mar 2025
p-ISSN: 2395-0072
www.irjet.net
Household Electricity Prediction using enhanced XGBoost Prasad Naidu.K1, Yamini.C2 1Student,Dept of MCA,KMM Inst. of P.G Studies,Tirupati,AP,India
2Assistant Professor, Dept of MCA,KMM Institute of P.G Studies,Tirupati,A.P,India
---------------------------------------------------------------------***--------------------------------------------------------------------
Abstract - Accurate household electricity consumption
integration is expected to reduce overfitting, improve model interpretability, and yield higher accuracy in forecasting household electricity usage patterns. Through extensive experimentation on real-world electricity consumption datasets, we evaluate the effectiveness of this hybrid model compared to traditional machine learning approaches.
forecasting is essential for efficient energy management and grid optimization. Traditional machine learning models, while effective, often struggle with feature redundancy and non-linear relationships in energy consumption data. This study proposes an integrated approach which enhances XGBoost which uses Rotation Forest to improve prediction accuracy. Rotation Forest enhances feature diversity by generating multiple diverse decision trees, while XGBoost efficiently captures complex patterns and interactions in the data. The hybrid model is evaluated on real-world household electricity consumption datasets, demonstrating superior predictive performance compared to standalone models. Experimental results indicate that the proposed method reduces forecasting errors and enhances generalization capability, making it a promising solution for smart grid applications.
This research contributes to the growing field of energy analytics by proposing a novel ensemble learning approach that combines feature transformation and boosting techniques to enhance electricity consumption prediction accuracy.
1.1 Rotation Forest Rotation Forest is an ensemble learning method introduced by Rodríguez et al. (2006) that enhances the diversity of base classifiers while maintaining high accuracy. It is particularly effective for classification tasks and has been widely applied in various domains, including pattern recognition, medical diagnosis, and energy consumption forecasting.
Key Words: machine learning, clustering, logistic regression, unsupervised learning, supervised learning, data analysis, prediction, classification, segmentation, data mining, model evaluation, algorithms, and applications
1.INTRODUCTION
The core idea of Rotation Forest is to increase feature diversity among base classifiers by applying principal component analysis (PCA) to random subsets of the input features. This transformation generates different feature representations for each base learner, ensuring that they learn distinct patterns. Unlike traditional ensemble methods such as Bagging or Boosting, where base classifiers may share similar decision boundaries, Rotation Forest fosters greater independence among them, leading to improved generalization performance.
Accurate household electricity consumption prediction is essential for efficient energy management, load balancing, and demand forecasting. Traditional machine learning models such as decision trees and ensemble methods like XGBoost have demonstrated strong predictive capabilities in this domain. However, further improvements in prediction accuracy and model robustness can be achieved by integrating advanced ensemble learning techniques. This study explores the integration of Rotation Forest with XGBoost to enhance the accuracy and generalization ability of electricity consumption prediction models. Rotation Forest, an ensemble learning technique based on feature transformation and individual classifier diversity, has shown promise in improving predictive performance. By leveraging XGBoost’s gradient boosting framework and Rotation Forest’s ability to enhance feature diversity, this hybrid approach aims to create a more powerful predictive model.
1. Ensemble of Decision Trees Rotation Forest constructs multiple decision trees as base classifiers, similar to Random Forest, but with a key difference in how features are transformed. 2. Feature Subset Extraction & PCA Rotation The feature set is randomly divided into smaller subsets. Principal Component Analysis (PCA) is applied to each subset separately. The transformed features are then used to train individual decision trees. This introduces diversity while preserving the original information.
The proposed methodology involves applying Rotation Forest to generate diverse feature subsets and using these transformed features as inputs for XGBoost. This
© 2025, IRJET
|
Impact Factor value: 8.315
|
ISO 9001:2008 Certified Journal
|
Page 1049