Domestic appliance power consumption measurement was, until recently, a problem without a satisfying solution. It required the use of a measuring device for each appliance to be studied, and thus the spending of a considerable amount of both money and time. The technological advancements made in the past few decades have enabled the engineering of smart devices that connect to the central panel of a building and log the features of the electrical current passing through it. Using Machine Learning algorithms, we can create models that extract individual appliance information (“signature”) from the signals recorded by these measuring devices. This process can lead to the production of systems that could be particularly useful for the consumers. They would not only allow individuals to alter their power consumption profile to minimise their spending and environmental impact, but also notify them if an appliance seems to be malfunctioning. In addition, the energy providers could harness the potential of usage statistics collected from their customers to estimate the energy demand for any given moment within a day. This would prevent the production of excess energy or the overloading of the power supply network infrastructure. The objective of this work is the implementation of an efficient Deep Neural Network (DNN) model that will be able to predict the state (On/Off) of a set of electrical appliances during a specific time span, based on the aggregate power signal of the house within which they operate. The contribution of this work concerns the use of a Recurrent Neural Network (RNN) that categorises the behaviour of multiple appliances (multi-label classification). The results are quite promising and pave the way for a more in-depth treatment of the problem.