Data mining-based inhabitant action predictor for smart homes using controlled synthetic data

Smart home research has led to the development of many sophisticated network protocols, smart appliances, and home gateway technologies. A smart home is a networked home containing various electrical and electronic devices controlled by a home gateway which manages the appliances and connects the home to service providers via the Internet. Smart homes are generally designed to assist people with cognitive impairments, seniors, and/or people with physical disabilities in their day-to-day activities. A key element in building such a user-adaptive smart home is to fully utilize the computational capabilities and automate the working of the smart appliances based on the inhabitants' appliance usage patterns.;The unavailability of real smart home data, due to cost and privacy issues, led us to design a synthetic data generator based on discrete-event simulation, capable of generating plausible spatial-scenario-based smart home data. We used a controlled variation technique to generate similar data repeatedly so it can be used to test the data mining application. We induced temporal heterogeneity to represent time variations in day-to-day user device interactions. We also used a parameterizable Discrete Time Markov Chain (DTMC) to generate varying proportions of patterned and non-patterned smart home data. We found that our prediction system gave a useful rate of correct predictions over a wide range of tuning parameters and proportions of patterned and non-patterned data.;We strongly believe that this system will be an important component of the basic prototype platform for promoting independence to seniors and/or the physically challenged, who require assisted living to remain in their own homes.;The goal of this thesis is to build a system which will assist device automation in smart homes based on the device usage patterns of a smart home inhabitant. By applying suitably adapted sequential data-mining techniques to historical smart home data, consisting of an inhabitant's device interactions, we extract device usage patterns that permit us to predict each user's next action. The predicted action could then be used to send signals to the appropriate devices through the home gateway, thereby automating the home.