| Title |
Validating an AI Model for Building Lifespan Prediction Using Big Data |
| DOI |
https://doi.org/10.5659/JAIK.2024.40.5.3 |
| Keywords |
Building lifespan; Big Data; Prediction model; Artificial Intelligence; Deep Neural Network |
| Abstract |
Accurately estimating a building's lifespan is crucial for assessing its asset value and determining its economic and environmental feasibility,
which is key for decision-making in the construction industry. However, because it's nearly impossible to precisely estimate the lifespan of
each building due to the wide range of influencing factors, most studies have used uniform lifespans based on the building's primary
structural type. To address this limitation, 1,812,700 records were analyzed of buildings constructed and demolished in Korea to predict each
building's lifespan with greater accuracy. Based on the previous study, a prediction model was developed using both deep learning and
traditional machine learning methods. This study evaluated whether the building lifespan prediction model experienced overfitting based on the
data period used to create the model. A performance evaluation was also conducted, comparing models using only key factors to those using
a broader set of factors. The results showed that among the machine learning models, the artificial neural network model, a nonlinear
approach, maintained high predictive accuracy without overfitting, regardless of the data period used. The model that used all available factors
performed better than those based on just a few key factors. This research demonstrates the viability of using big data and AI for building
lifespan prediction, providing a more reliable method for estimating building lifespan tailored to each building's unique characteristics. This
approach meets a growing societal demand for more accurate building lifespan predictions. |