Delivering analytics services for smart homes

Smart homes empowered with Internet of Things aim at providing enhanced user comfort by means of non-intrusive home automation systems. Unfortunately, smart homes today have not yet reached the required level of maturity. Current smart home inhabitants struggle with complex user interfaces and static appliance configurations. The smart home user interface needs to go beyond traditional tools such as keyboards and touchpads for better usability. Another limitation of current smart homes is that, they are not intelligent enough to change the status of home appliances based on usage history. On the other hand, recommending appropriate device settings to users based on their past behavior is also a big challenge. These challenges need to be properly addressed in order to bring sensible home automation solutions to the mass market. This paper presents design and development of the techniques that enable smart human-device interfaces and an appliance usage-prediction engine to aid home automation systems. We also present a recommendation system designed especially for smart homes. The appliance usage-prediction engine predicts the status of devices at a given time. Initially, the random forest and gradient boosting methods are used to train our appliance usage-prediction engine. Our highly accurate ensemble model is then developed using random forest and gradient boosting methods as basis. In addition, we have used real data to verify the functionality of our ensemble models and achieved significantly high accuracy levels of around 90% in trial runs. Our proposed system also demonstrates successful prediction of both device status at a given time of the day as well as recommendations based on user behavior.