Computational intelligence techniques for missing data imputation

摘要

Despite considerable advances in missing data imputation techniques over the last three decades, theudproblem of missing data remains largely unsolved. Many techniques have emerged in the literatureudas candidate solutions, including the Expectation Maximisation (EM), and the combination of autoassociativeudneural networks and genetic algorithms (NN-GA). The merits of both these techniquesudhave been discussed at length in the literature, but have never been compared to each other. Thisudthesis contributes to knowledge by firstly, conducting a comparative study of these two techniques..udThe significance of the difference in performance of the methods is presented. Secondly, predictiveudanalysis methods suitable for the missing data problem are presented. The predictive analysis inudthis problem is aimed at determining if data in question are predictable and hence, to help inudchoosing the estimation techniques accordingly. Thirdly, a novel treatment of missing data for onlineudcondition monitoring problems is presented. An ensemble of three autoencoders together withudhybrid Genetic Algorithms (GA) and fast simulated annealing was used to approximate missinguddata. Several significant insights were deduced from the simulation results. It was deduced that forudthe problem of missing data using computational intelligence approaches, the choice of optimisationudmethods plays a significant role in prediction. Although, it was observed that hybrid GA and FastudSimulated Annealing (FSA) can converge to the same search space and to almost the same valuesudthey differ significantly in duration. This unique contribution has demonstrated that a particularudinterest has to be paid to the choice of optimisation techniques and their decision boundaries.udiiiudAnother unique contribution of this work was not only to demonstrate that a dynamic programmingudis applicable in the problem of missing data, but to also show that it is efficient in addressing theudproblem of missing data. An NN-GA model was built to impute missing data, using the principleudof dynamic programing. This approach makes it possible to modularise the problem of missinguddata, for maximum efficiency. With the advancements in parallel computing, various modules ofudthe problem could be solved by different processors, working together in parallel. Furthermore, audmethod for imputing missing data in non-stationary time series data that learns incrementally evenudwhen there is a concept drift is proposed. This method works by measuring the heteroskedasticityudto detect concept drift and explores an online learning technique. New direction for research, whereudmissing data can be estimated for nonstationary applications are opened by the introduction of thisudnovel method. Thus, this thesis has uniquely opened the doors of research to this area. Manyudother methods need to be developed so that they can be compared to the unique existing approachudproposed in this thesis.udAnother novel technique for dealing with missing data for on-line condition monitoring problem wasudalso presented and studied. The problem of classifying in the presence of missing data was addressed,udwhere no attempts are made to recover the missing values. The problem domain was then extendedudto regression. The proposed technique performs better than the NN-GA approach, both in accuracyudand time efficiency during testing. The advantage of the proposed technique is that it eliminatesudthe need for finding the best estimate of the data, and hence, saves time. Lastly, instead of usingudcomplicated techniques to estimate missing values, an imputation approach based on rough sets isudexplored. Empirical results obtained using both real and synthetic data are given and they provide audvaluable and promising insight to the problem of missing data. The work, has significantly confirmedudthat rough sets can be reliable for missing data estimation in larger and real databases.