Desmira (2026) Enhancement of prediction models using a modified feature selection method in energy consumption. Doctoral thesis, Universiti Teknikal Malaysia Melaka.
![[img]](http://eprints.utem.edu.my/style/images/fileicons/text.png) |
Text (Full Text)
Enhancement of prediction models using a modified feature selection method in energy consumption.pdf - Submitted Version Restricted to Registered users only Download (4MB) |
Abstract
Enhancing energy efficiency can encourage customers to adopt a more logical approach when consuming electricity. Hence, it is imperative to devise solutions that address the challenge of forecasting and meeting future energy requirements, particularly when employing machine learning algorithms to analyze data. The issue of using machine learning to choose features for projecting global energy consumption has not yielded much effective computational times. Therefore, the need for feature selection to reduce data for predicting energy consumption yields beneficial results, namely, reducing the costs associated with energy-consuming equipment. However, several problems were identified such as computation time and efficiency, data transfer and dimensionality, feature selection limitations and inefficiencies, whereas analyzing current machine learning features reveals substantial constraints in predicting room energy use. The proposed solution is to add one stage of filtering with data grouping between the combination of PCA and BGA feature selection., focusing on improving computational efficiency within energy consumption prediction models for ANN and ANFIS. Despite these efforts, feature selection still presents accuracy issues that require further exploration and development. The objective of this research is to address the challenges and limitations associated with machine learning in predicting energy. The three main objectives of this research are as follows: adding a filtering stage after PCA feature selection, develop data grouping and combine feature selection, and evaluate the performance of combined PCA and BGA methods. In this study, a total of 3350 data collected based on real retrieval in the PVTE UNTIRTA laboratory for 3 months are used. This dataset is divided into 70% training, 20% testing, and 10% prediction datasets and compares the performances of the energy prediction models before and after normalization for prediction. The application of PCA and BGA with filtering features selection in predicting the energy of a building has contributed significantly. It can be seen that there is a reduction in the computational speed of almost 99,99 % of the tested models, namely the ANN and ANFIS models. PCA and BGA feature selection with filtering variable reduction input can increase efficiency by reducing high-dimensional data, making it easier to use for predictions with machine learning. The results of this research, through the reduction of input variables and faster computational time, have the potential to generate a significant impact on various aspects of life, particularly in the industrial sector. This study will also optimize and improve the efficiency of energy usage, especially in the management of energy generation and consumption within buildings. Another benefit lies in the implementation and development of intelligent sensors for energy management in the residential sector. Furthermore, this research makes a substantial contribution to the achievement of the Sustainable Development Goals (SDGs), particularly in relation to clean and affordable energy.
| Item Type: | Thesis (Doctoral) |
|---|---|
| Uncontrolled Keywords: | PCA and BGA feature selection |
| Subjects: | T Technology T Technology > TK Electrical engineering. Electronics Nuclear engineering |
| Divisions: | Library > Tesis > FTKE |
| Depositing User: | Norhairol Khalid |
| Date Deposited: | 07 May 2026 02:05 |
| Last Modified: | 07 May 2026 02:05 |
| URI: | http://eprints.utem.edu.my/id/eprint/29911 |
| Statistic Details: | View Download Statistic |
Actions (login required)
 |
View Item |