Use this URL to cite or link to this record in EThOS:
Title: Automated theory selection using agent based models
Author: Stratton, Robert James
ISNI:       0000 0004 5368 3167
Awarding Body: King's College London
Current Institution: King's College London (University of London)
Date of Award: 2015
Availability of Full Text:
Access from EThOS:
Access from Institution:
Models are used as a tool for theory induction and decision making in many contexts, including complex and dynamic commercial environments. New technological and social developments — such as the increasing availability of real-time transactional data and the rising use of online social networks — create a trend towards modelling process automation, and a demand for models that can help decision making in the context of social interaction in the target process. There is often no obvious specification for the form that a particular model should take, and some kind of selection procedure is necessary that can evaluate the properties of a model and its associated theoretical implications. Automated theory selection has already proven successful for identifying model specifications in equation based modelling (EBM), but there has been little progress in developing automatic approaches to agent based model (ABM) selection. I analyse some of the automation methods currently used in EBM and consider what innovations would be required to create an automated ABM specification system. I then compare the effectiveness of simple automatically specified ABM and EBM approaches in selecting optimal strategies in a series of encounters between artificial corporations, mediated through a simulated market environment. I find that as the level of interaction increases, agent based models are more successful than equation based methods in identifying optimal decisions. I then propose a fuller framework for automated ABM model specification, based around an agent-centric theory representation which incorporates emergent features, a model-to-theory mapping protocol, a set of theory evaluation methods, a search procedure, and a simple recommendation system. I evaluate the approach using empirical data collected at two different levels of aggregation. Using macro level data, I derive a theory that represents the dynamics of an online social networking site, in which the data generating process involves interaction between users, and derive management recommendations. Then, using micro level data, I develop a model using individual-level transaction data and making use of existing statistical techniques — hidden Markov and multinomial discrete choice models. I find that the results at both micro and macro level offer insights in terms of understanding the interrelationship between exogenous factors, agent behaviours, and emergent features. From a quantitative perspective, the automated ABM approach shows small but consistent improvements in fit to the target empirical data compared with EBM approaches.
Supervisor: McBurney, Peter John ; Luck, Michael Mordechai Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available