Novel Methods for Human Activity Prediction in Ambient Intelligence and Smart Environments Using Auditory and Contextual Data

Abstract

Ambient Intelligence and smart environments are a leading area for computer science research, and have the potential to radically change everyday lives in the home, workplace, and healthcare settings. In this thesis I investigate the components that make up an Ambient Intelligence system, determine if it is possible to create a smart environment predictive model using only audio of a person’s activity as the data input, and ascertain whether the application of time-windowed contextual data in a smart environment model improves the accuracy of next activity prediction. Several research contributions are made including a narrative review of Ambient Intelligence, a systematic literature review of computational methods of human behaviour prediction, a method for improved audio classification using ensemble machine learning models, and two proofs-of-concept in smart environment activity prediction. One which demonstrates prediction of an occupant’s behaviour performed using only environmental sounds, and another which demonstrates how to improve prediction accuracy by using contextual data coupled with segmenting of activities into time-windows. This thesis follows the Thesis by Journal format and comprises five research papers and one conference paper, which were produced over the entire course of study.

Date of Award	1 Aug 2023
Original language	English
Awarding Institution	The University of Manchester
Supervisor	David Morris (Supervisor) & Simon Harper (Supervisor)