| id |
ecaade2023_31 |
| authors |
Canli, Ilkim, Gursel Dino, Ipek and Kalkan, Sinan |
| year |
2023 |
| title |
Useful Daylight Illuminance Prediction Under Data Imbalance in an Urban Context |
| source |
Dokonal, W, Hirschberg, U and Wurzer, G (eds.), Digital Design Reconsidered - Proceedings of the 41st Conference on Education and Research in Computer Aided Architectural Design in Europe (eCAADe 2023) - Volume 2, Graz, 20-22 September 2023, pp. 599–608 |
| doi |
https://doi.org/10.52842/conf.ecaade.2023.2.599
|
| summary |
Optimal daylight illumination can aid sustainable design by improving occupants’ psychological and physical health, visual and thermal comfort and decreasing electrical lighting energy usage in buildings. However, dense urban areas can result in restricted daylight access in buildings. Therefore, daylight analysis considering surrounding buildings is important for implementing daylighting strategies. Useful Daylight Illuminance (UDI) is a performance metric that can quantify the annual illuminance levels within certain illumination classes (UDIfell-short, UDIsupplementary, UDIautonomous, and UDIexceeded). UDI can be predicted using machine-learning (ML) methods. However, the calculated data is typically unevenly distributed, generally following a power-law distribution, which causes ML models to underperform for UDI classes with less data. Simulations can be utilized to increase the less dispersed data in the dataset; however, at the urban scale, the computational cost of collecting simulation data for daylighting analysis makes it difficult to augment data with simulations. To undertake this challenge, in this study, SMOTE (Synthetic Minority Oversampling Technique) was applied to augment data to increase the prediction performance of the ML model. The results showed that augmenting the data in the classes which are unevenly distributed leads to an increase in ML model prediction performance. This method shows that SMOTE can be used to increase the performance of ML models during UDI estimation at the urban scale. |
| keywords |
Daylight Illumination, Machine Learning Prediction, Useful Daylight Illuminance, Data Imbalance |
| series |
eCAADe |
| email |
|
| full text |
 file.pdf (686,656 bytes) |
| references |
Content-type: text/plain
|
| last changed |
2023/12/10 10:49 |
|
