Graduation Semester and Year
2018
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Industrial Engineering
Department
Industrial and Manufacturing Systems Engineering
First Advisor
Shouyi Wang
Abstract
The main purpose of this study is feature engineering/learning from multivariate (MV) time-series to achieve a more interpretable model by dimension reduction. This aim is fulfilled in 2 main parts. In part 1, we proposed a network estimation approach namely SWDN which stands for sparse weighted directed network. In this approach, the directed subgraph of the underlying network was detected by maximum spanning tree (MST) algorithm that created a null model of connections with maximum inter-dependence (pairwise correlation or mutual information) forming the backbone structure of the MV time-series as an empirical reference. The edge weights were estimated using the linear conditional Gaussian parameters with the maximum likelihood. The efficiency of the proposed method (SWDN) was evaluated on the publicly available simulated fMRI data-set generated based on BOLD with different simulation parameters and in comparison with other network construction methods, it was verified to outperform Granger and lag-based methods under some circumstances. We applied SWDN as a feature extraction tool, and classified Parkinson's Disease (PD) fMRI data by finding the discriminative patterns between estimated network of PDs vs controls and achieved 75 % test accuracy via N-fold cross-validation. In part 2, we made an extensive feature analysis framework for MV time-series. This framework consisted of extensive features extraction, post processing and a novel proposed feature selection technique based on mutual information and sparsity learning with embedded group structure. The multivariate time-series in this part was functional near infrared spectroscopy (fNIRS) which is a noninvasive neuroimaging technique for brain activity monitoring. We applied the proposed supervised extensive sparse feature learning method on two data-sets to extract and select features and by applying machine learning and data mining approach and algorithms to classify participants with brain disorder/disease from the controls.
Keywords
Multivariate time series analysis, Sparse weighted directed network, Mutual information, fNIRS, Biomarkers, fMRI
Disciplines
Engineering | Operations Research, Systems Engineering and Industrial Engineering
License
This work is licensed under a Creative Commons Attribution-NonCommercial-Share Alike 4.0 International License.
Recommended Citation
Hosseini, Rahilsadat, "Proposing a Sparse Weighted Directed Network Construction Method and a Novel Mutual-Information based Sparse Feature Selection Algorithm for Multivariate Time-series Analysis and its Application in Medical Diagnostic Problems" (2018). Industrial, Manufacturing, and Systems Engineering Dissertations. 175.
https://mavmatrix.uta.edu/industrialmanusys_dissertations/175
Comments
Degree granted by The University of Texas at Arlington