Natural ventilation efficiency is a sustainable solution for improving indoor environment quality, especially in older structures like churches. Traditional computational tools for evaluating natural ventilation potential are often time-consuming, labor-intensive, and resource-demanding. This thesis, titled "Prediction of Average Wind Pressure Coefficient on Building Surfaces by Artificial Neural Network," presents an efficient, fast, and accurate tool for predicting the average wind pressure coefficient with consideration of terrain complexity and variety of building dimensions, a critical parameter for assessing natural ventilation potential in buildings.

In this research, Valbo church, which is located in Gästrikland, Sweden, is chosen as the baseline and in order to create a data base containing different development for this kind of church, we tried to build 378 cases based on diminishing and enlarging the size of the Valbo church.

The proposed tool leverages a multilayer perceptron (MLP) network, trained using PyTorch with Optuna for hyperparameter optimization and k-fold cross-validation to ensure robustness and generalization. An additional test group was employed to validate the model's reliability and accuracy, achieving promising results. By using this tool, the prediction result on Valbo church is quite satisfied, the worst result is more than 0.98 in correlation parameter (R). In other cases where the dimension values can be found among the ones in the database, the worst result is still more than 0.97 in R. 

This study not only demonstrates the effectiveness of artificial neural networks (ANNs) in predicting average wind pressure coefficients but also highlights their broader potential in building technology applications. With further research, ANN-based tools could extend their capabilities to predict detailed wind pressure coefficients distribution and flow patterns, offering significant advancements in the evaluation and design of natural ventilation systems.