Gradient Boosting Regression Tree with Particle Swarm Optimization Framework for Architectural Design

From the past few years, architectural design for buildings has become essential for achieving improved structural efficiency, energy performance and spatial functionality. However, traditional models often failed in capturing nonlinear and complex relationships among design variables and performance metrics. For addressing this issue, a novel optimization framework which combines Gradient Boosting Regression Tree (GBRT) with Particle Swarm Optimization (PSO) is proposed in this research for high-accuracy prediction in architectural applications. Initially, data is collected from structural and architectural design dataset that comprising mid-rise reinforced concrete buildings with detailed spatial as well as performance parameters. This dataset is preprocessed by using normalization to standardize feature ranges in specific range. Then, feature engineering is done by utilizing Principal Component Analysis (PCA) that transforms correlated features into uncorrelated components by reducing dimensionality and improving model efficiency. Next, GBRT is utilized for predicting architectural performance by modeling complex nonlinear relationships by offering reduced overfitting. Finally, PSO is used to fine-tune the hyperparameters of GBRT model by ensuring optimal predictive performance. The proposed GBRT based PSO had achieved better results in terms of Mean Square Error (MSE) of 0.00002 when compared with existing Artificial Neural Network-PSO (ANN-PSO) model.