Novel Automatic Feature Engineering for Carbon Emissions Prediction Base on Deep Learning

Lee, Zne-Jung; Lin, Yun; Yang, Zhen‐Biao; Chen, Zhongyuan; Fan, Wei‐Qiang; Lee, Chen-Hsin

doi:10.1109/ecei57668.2023.10105367

Cited by 1 publication

(1 citation statement)

References 10 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…However, the majority of research focuses on machine learning for predicting carbon emissions, whereas feature extraction for predicting carbon emissions receives less attention. To estimate carbon emissions, Lee et al [15] proposed using deep learning with autonomous feature engineering; however, parameters still need to be fine-tuned to increase prediction accuracy. In this research, we propose a DNN-based building carbon emissions prediction with improved PSO.…”

Section: Introductionmentioning

confidence: 99%

Building Carbon Emissions Prediction Based on Deep Learning Network with Improved Particle Swarm Optimization

Chai,

Lin,

Wang

et al. 2023

Artificial Intelligence Evolution

View full text Add to dashboard Cite

Buildings' carbon emissions are the main contributor to climate change. The world needs to be able to foresee and further reduce construction carbon emissions if it wants to prevent the worst effects of climate change. The main challenge in carbon emission prediction for buildings is how to increase algorithm accuracy. Therefore, a novel technique for calculating carbon emissions is proposed in this study. The proposed technique uses improved particle swarm optimization (PSO) and deep neural network (DNN) to anticipate carbon emissions. DNN is employed in the proposed technique to forecast carbon emissions. The parameters used are fine-tuned using the improved PSO. Additionally, it chooses features that increase the predictability of carbon emissions. In this study, many methods for predicting carbon emissions are examined, including decision tree (DT), random forest (RF), support vector regression (SVR), DNN, and the proposed method. The outcomes demonstrate that the proposed technique can lower root mean square logarithmic error (RMSLE) prediction, and the final testing result of RMLSE is 0.3476 in this study. It demonstrates that the proposed method performs better than other alternatives when compared and has good implementation ability.

show abstract

Section: Introductionmentioning

confidence: 99%