Use this URL to cite or link to this record in EThOS: https://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.769240
Title: New approaches for clustering time series data
Author: Bakoben, Maha
ISNI:       0000 0004 7656 8751
Awarding Body: Imperial College London
Current Institution: Imperial College London
Date of Award: 2016
Availability of Full Text:
Access from EThOS:
Access from Institution:
Abstract:
Time series data describe features of objects sequentially in time. The analysis of this data is useful in understanding the past and estimating the future. One of the promising methods in analysis of such data is cluster analysis which aims at finding underlying similarities, such that objects can be partitioned into clusters. Diverse implementations of this methodology are available. For example, clustering gene expression profiles in biology; clustering stock time series in finance; clustering daily temperature in meteorology and many others. Unlike standard data, the intrinsic features of the time series need to be considered in the analysis, such as serial dependence. This thesis takes a statistical approach in which the dynamic features are explained using appropriate time series models. The focus will be directed toward clustering of statistical model parameters. However, the inevitable uncertainty associated with parameter estimates is a critical issue that might negatively affect clustering in the parameter space. One of the main contributions of this thesis is the development of a new clustering approach that addresses this issue. Rigorous studies of the properties and performance of the new approach are provided. In addition, the computation of the proposed approach relies on Monte Carlo sampling, hence novel procedures are developed for this purpose. The time series clustering research is motivated and fully developed by a real data application of credit card behaviours. The problem is studied in the behavioural credit scorecards framework in which clustering of the credit account behaviours might assist in identifying distinct risk groups. We explore through logistic regression models the relationships between the behaviour clusters and the probability of default events.
Supervisor: Bellotti, Anthony Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID: uk.bl.ethos.769240  DOI:
Share: