Hybrid Explainable Smart House Control System

Abstract

This paper describes hybrid method for smart house control system, where control is made by training neural network with user habit data and then applying fuzzy rule set and computing with words engine to provide user with explanation why control change to light, heating or ventilation was made. For neural networks two different approaches were tested: classical support vector machine and newly emerged ML.Net framework. Both methods could correctly predict user desired controls with ~99% accuracy, but ML.Net was recommended for further use because it was less selective for the data format, on the other hand SVM required data separation for binary classification thus increasing required number of SVM machines if control areas would expand. Given verbal explanations accurately described current house control situation to the users, thus increasing confidence level for Explainable Artificial Intelligence.