Validation of Neural Network Model for Predicting Airtightness of Residential and Non-residential Units in Poland

Abstract This paper presents validation of the neural network model for predicting airtightness of residential and non-residential units. Proposed model developed in previous research utilizes a neural network in prediction of airtightness and it is obtained based on in situ measurements at 58 units in Croatia. Model applicability earlier was tested by independent validation through 20 additional measurements in Croatia and on 5 measurements in Serbia. This paper presents validation of neural network model for predicting airtightness of residential and non-residential units on database formed beyond the regional area. Database used for model validation in this paper consists of 20 residential and non-residential units from Poland with building construction technology similar to Croatian ones. Comparison of the results obtained by measurements and prediction model indicates that the model gives consistent results on the validation data set with similar building construction technology and demonstrates that model is not locally conditioned.