An Unsupervised Neural Model to Analyse Thermal Properties of Construction Materials

This study proposes a new approach to feature selection and the identification of underlaying factors. The goal of this method is to visualize and extract information from complex and high dimensional data sets. The model proposed is an extension of Maximum Likelihood Hebbian Learning [14], [5], [15] based on a family of cost functions, which maximizes the likelihood of identifying a specific distribution in the data while minimizing the effect of outliers [7], [10]. We demonstrate a hierarchical extension method which provides an interactive method for identifying possibly hidden structure in the dataset. We have applied this method to study the thermal evolution of several construction materials under different thermal and humidity environmental conditions.