Development of a machine learning-based framework for predicting failures in heat supply networks