Time Series Prediction Using Recurrent Neural Networks And Boosting: An Experimental Study In Pharmaceutical Product Formulation

Time Series Prediction Using Recurrent Neural Networks And Boosting: An Experimental Study In Pharmaceutical Product Formulation

Simple item page

dc.contributor.author	Wei Yee, Goh
dc.date.accessioned	2016-10-24T03:31:43Z
dc.date.available	2016-10-24T03:31:43Z
dc.date.issued	2002-05
dc.description.abstract	This thesis is devoted to the development of Artificial Neural Network (ANN) techniques for solving time-series prediction problems. The research is focused on the use of recurrent neural networks for devising a comprehensible framework for pharmaceutical product formulation using time series prediction approach. In particular, the framework explores the learning paradigms of ANNs for conducting the experimental design and analysis. Based upon existing methodologies, novel ANN architectures are proposed for time series analyses in the process of pharmac~utical product formulation. The Elman recurrent neural network is employed for the prediction of in-vitro dissolution profiles of matrix-controlled-release theophylline pellets preparations. Feedback links in this network perform recursive computation, and provide the ability of dynamical representation of information in time series analyses, especially for nonlinear dynamical systems. The experimental results have successfully demonstrate¢ the potentials of the Elman-based networks for formulating pharmaceutical products to meet the desired drug release characteristics. Furthermore, reliability of the network performance is statistically assessed using the bootstrap method. The theoretical foundations and operational strategies of boosting have been elaborated in details. The standard AdaBoost algorithm that is often employed in classification tasks is modified to accommodate multiple Elman networks for time-series prediction problems. Several simulation studies are conducted using benchmark data sets, and the results are compared with those obtained from other published methods. The results indicate that multiple Elman networks coupled with the modified AdaBoost algorithm are capable of improving generalisation of individual networks. In addition, the pharmaceutical formulation problems are re-visited to assess the practical applicability of multiple predictors devi~ed based on the Elman networks. The results are evaluated statistically using hypotheses tests to justify the effectiveness of the proposed system. In summary, this research work has revealed the benefits of using multiple Elman-based networks with boosting as a unified and convenient framework for utilizing techniques in time series prediction for pharmaceutical product formulation.	en_US
dc.identifier.uri	http://hdl.handle.net/123456789/2808
dc.subject	Computer	en_US
dc.subject	Networks And Boosting	en_US
dc.title	Time Series Prediction Using Recurrent Neural Networks And Boosting: An Experimental Study In Pharmaceutical Product Formulation	en_US
dc.type	Thesis	en_US

Files

License bundle

Now showing 1 - 1 of 1

Name:: license.txt
Size:: 1.71 KB
Format:: Item-specific license agreed upon to submission
Description:

Collections

Pusat Pengajian Sains Komputer - Tesis