Date of Award

5-2015

Document Type

Master Thesis

Degree Name

Masters of Science (Research)

Department

Computing

First Advisor

Dr Robert Sheehy

Second Advisor

Mr Andrew Shields

Third Advisor

Dr Pat Doody

Abstract

In this thesis, the application of Artificial Intelligence (A.I.) techniques are examined in order to further develop and improve the field of Reality Mining. Reality Mining refers to the collection of machine sensed environmental data, from which, it is hoped future human social behaviour can be predicted. Large datasets for Reality Mining study have been generated by recording the patterns of human activity and social interactions over an extended period of time. Reality Mining techniques can provide information on important social clues for the better understanding of how humans live their lives. This information can then be used to help in the development of real world applications for example patient diagnosis, better transportation facilities, traffic management and improvements in healthcare systems. The ubiquitous nature of mobile phones, coupled with their ever increasing sophistication, means they have become an ideal tool for collecting such large datasets. Reality Mining techniques can then be applied to find the predictable patterns of human behaviour and identify relationships between individuals.

In this thesis, Machine Learning Algorithms developed within A.I. are introduced as predictive analysis models for Reality Mining. These models are applied on individuals past location patterns to predict its future locations. This thesis shows that these models can predict the future location of an individual once the past locations of that individual are known. These patterns are observed from anonymised datasets collected using mobile phones in Reality Mining experiment. The Reality Mining experiment conducted in 2004 by Massachusetts Institute of Technology (MIT) Media Laboratory represents approximately 500,000 hours of data on user’s location, communication, and device usage behaviour. This data was collected over nine months by monitoring the mobile phone usage of one hundred participants. The dataset was then anonymised and made publicly available for academic research purposes. The predictive models are then compared by measuring performance in predicting unseen test data and the time taken to train the model in question.

The location pattern of people may highly vary from one to another depending on their random movement behaviour. It is believed that this affects in predicting location of an individual. Shannon Entropy is used to measure this randomness level. With this knowledge, the affects in accuracy of location prediction by predictive models is experimented.

Access Level

info:eu-repo/semantics/openAccess

Coverage

July 2024

Download

Share

COinS