Mould threat of building envelopes classified by unsupervised and supervised machine learning methods analysing multidimensional signals from gas sensors

Abstract

Mould thread of buildings appears when the moisture content of envelopes increases, and is a significant problem in building operation. This problem is important in terms of architecture and building construction, residents’ health as well as visual reasons. There are many methods of evaluating mould infestation – traditional biological (mycological), molecular microbiological (Polymerase Chain Reaction), and chemical (chromatography) techniques. One of the new and early detection methods is appliance of gas sensors arrays, which together with appropriate data analysis algorithm form an electronic nose. The important issue connected with correct of an e-nose functioning is application of the proper model enabling visualization and interpretation of the raw data – multidimensional signals from gas sensors. In this work are presented examples of unsupervised and supervised machine learning methods for analysis of multidimensional readouts form MOS sensor matrix. Developed procedure allow showing which observation would be assigned to clusters which are connected with condition of buildings and their level of mould thread.