This thesis describes the research undertaken for the degree of Doctor of Philosophy, testing the hypothesis that spectrometer data can be used to establish usable relationships for prediction of pasture quality attributes. The research data consisted of reflectance measurements of various temperate pasture types recorded at four different times (years 2000 to 2002), recorded by three hyperspectral sensors, the in situ ASD, the airborne HyMap and the satellite-borne Hyperion. Corresponding ground-based pasture samples were analysed for content of chlorophyll, water, crude protein, digestibility, lignin and cellulose at three study sites in rural Victoria, Australia. This context was used to evaluate effects of sensor differences, data processing and enhancement, analytical methods and sample variability on the predictive capacity of derived prediction models. Although hyperspectral data analysis is being applied in many areas very few studies on temperate pastures have been conducted and hardly any encompass the variability and heterogeneity of these southern Australian examples. The research into the relationship between the spectrometer data and pasture quality attribute assays was designed using knowledge gained from assessment of other hyperspectral remote sensing and near-infrared spectroscopy research, including bio-chemical and physical properties of pastures, as well as practical issues of the grazing industries and carbon cycling/modelling. Processing and enhancement of the spectral data followed methods used by other hyperspectral researchers with modifications deemed essential to produce better relationships with pasture assay data. As many different methods are in use for the analysis of hyperspectral data several alternative approaches were investigated and evaluated to determine reliability, robustness and suitability for retrieval of temperate pasture quality attributes. The analyses employed included stepwise multiple linear regression (SMLR) and partial least squares regression (PLSR). The research showed that the spectral research data had a higher potential to be used for prediction of crude protein and digestibility than for the plant fibres lignin and cellulose. Spectral transformation such as continuum removal and derivatives enhanced the results. By using a modified approach based on sample subsets identified by a matrix of subjective bio-physical and ancillary data parameters, the performance of the models were enhanced. Prediction models from PLSR developed on ASD in situ spectral data, HyMap airborne imagery and Hyperion and corresponding pasture assays showed potential for predicting the two important pasture quality attributes crude protein and digestibility in hyperspectral imagery at a few quantised levels corresponding to levels currently used in commercial feed testing. It was concluded that imaging spectrometry has potential to offer synoptic, simultaneous and spatially continuous information valuable to feed based enterprises in temperate Victoria. The thesis provide a significant contribution to the field of hyperspectral remote sensing and good guidance for future hyperspectral researchers embarking on similar tasks. As the research is based on temperate pastures in Victoria, Australia, which are dominated by northern hemisphere species, the findings should be applicable to analysis of temperate pastures elsewhere, for example in Western Australia, New Zealand, South Africa, North America, Europe and northern Asia (China).

<p>The paper describes predictive modelling of biomarker data stemming from patients suffering from multiple sclerosis. Improvements of multivariate analyses of the data are investigated with the goal of increasing the capability to assign samples to correct subgroups from the data alone.</p><p>The effects of different preceding scalings of the data are investigated and combinations of multivariate modelling methods and variable selection methods are evaluated. Attempts at merging the predictive capabilities of the method combinations through voting-procedures are made. A technique for improving the result of PLS-modelling, called bagging, is evaluated.</p><p>The best methods of multivariate analysis of the ones tried are found to be Partial least squares (PLS) and Support vector machines (SVM). It is concluded that the scaling have little effect on the prediction performance for most methods. The method combinations have interesting properties – the default variable selections of the multivariate methods are not always the best. Bagging improves performance, but at a high cost. No reasons for drastically changing the work flows of the biomarker data analysis are found, but slight improvements are possible. Further research is needed.</p>

Problem: Real-time process and destructive test data were collected from a wood composite manufacturer in the U.S. to develop real-time predictive models of two key strength properties (Modulus of Rupture (MOR) and Internal Bound (IB)) of a wood composite manufacturing process. Sensor malfunction and data “send/retrieval” problems lead to null fields in the company’s data warehouse which resulted in information loss. Many manufacturers attempt to build accurate predictive models excluding entire records with null fields or using summary statistics such as mean or median in place of the null field. However, predictive model errors in validation may be higher in the presence of information loss. In addition, the selection of predictive modeling methods poses another challenge to many wood composite manufacturers. Approach: This thesis consists of two parts addressing above issues: 1) how to improve data quality using missing data imputation; 2) what predictive modeling method is better in terms of prediction precision (measured by root mean square error or RMSE). The first part summarizes an application of missing data imputation methods in predictive modeling. After variable selection, two missing data imputation methods were selected after comparing six possible methods. Predictive models of imputed data were developed using partial least squares regression (PLSR) and compared with models of non-imputed data using ten-fold cross-validation. Root mean square error of prediction (RMSEP) and normalized RMSEP (NRMSEP) were calculated. The second presents a series of comparisons among four predictive modeling methods using imputed data without variable selection. Results: The first part concludes that expectation-maximization (EM) algorithm and multiple imputation (MI) using Markov Chain Monte Carlo (MCMC) simulation achieved more precise results. Predictive models based on imputed datasets generated more precise prediction results (average NRMSEP of 5.8% for model of MOR model and 7.2% for model of IB) than models of non-imputed datasets (average NRMSEP of 6.3% for model of MOR and 8.1% for model of IB). The second part finds that Bayesian Additive Regression Tree (BART) produced most precise prediction results (average NRMSEP of 7.7% for MOR model and 8.6% for IB model) than other three models: PLSR, LASSO, and Adaptive LASSO.

Stimulus-driven changes in the gain of sensory neurons are well-documented, but relatively little is known about whether analogous gain-control can also be effected in a top-down manner. A recent psychophysical study demonstrated that sensitivity to luminance contrast can be modulated by a priori knowledge (de la Rosa et al., in press). In the present study, event-related potentials were used to resolve the stages of information processing that facilitate such knowledge-driven adjustments. Groupwise independent component analysis identified two robust spatiotemporal patterns of endogenous brain activity that captured experimental effects. The first pattern was associated with obligatory processing of contextual information, while the second pattern was associated with selective initiation of contrast gain adjustment. These data suggest that knowledge-driven contrast gain control is mediated by multiple independent electrogenic sources.

Stimulus-driven changes in the gain of sensory neurons are well-documented, but relatively little is known about whether analogous gain-control can also be effected in a top-down manner. A recent psychophysical study demonstrated that sensitivity to luminance contrast can be modulated by a priori knowledge (de la Rosa et al., in press). In the present study, event-related potentials were used to resolve the stages of information processing that facilitate such knowledge-driven adjustments. Groupwise independent component analysis identified two robust spatiotemporal patterns of endogenous brain activity that captured experimental effects. The first pattern was associated with obligatory processing of contextual information, while the second pattern was associated with selective initiation of contrast gain adjustment. These data suggest that knowledge-driven contrast gain control is mediated by multiple independent electrogenic sources.

Machine learning methods have been promising tools in science and engineering fields. The use of these methods in chemistry and drug design has advanced after 1990s. In this study, molecular electrostatic potential (MEP) surfaces of PCP-like compounds are modelled and visualized in order to extract features which will be used in predicting binding affinity. In modelling, Cartesian coordinates of MEP surface points are mapped onto a spherical self-organizing map. Resulting maps are visualized by using values of electrostatic potential. These values also provide features for prediction system. Support vector machines and partial least squares method are used for predicting binding affinity of compounds, and results are compared.

The e-government phenomena has become more important with the ever increasing number of implementations world wide. A model explaining the e-government adoption and the related measurement instrument a survey had been developed and validated in this study. In a post technology acceptance model (TAM) approach, theory of planned behavior (TPB) was extended to t the requirements of e-government context. The adoption of student loans service of the higher education student loans and accommodation association (KYK) was investigated to obtain data for empirical validation. The instrument was administered to over four-hundred students and partial least squares path modeling was employed to analyze the data. The results indicate that the model was an improvement over TAM in terms of predictive power. The constructs investigated in the study successfully explained the intention to use an e-government service.

Through the rapid expansion of information technologies, Learning Management Systems have become one of the most important innovations for delivery of education. Successful implementation and management of these systems are primarily based on the instructors&#039 / adoption. However, too few researches have been conducted to evaluate instructors&rsquo / adoption towards e-learning system as taking higher education as base. This study aims to understand behavioral intentions of higher education instructors towards Learning Management Systems and further to identify the influencing factors. A research model has been proposed based on the belief variables of the Technology Acceptance Model. Additionally, Application Characteristics, Individual, Social and Technological dimensions were considered to identify the effects of key variables on behavioral intention of users. A survey instrument has been developed and conducted with 224 academicians after a pilot study through its reliability and validity has been assured. Although the items of the survey instrument were based on the literature, an explanatory factor analysis was performed to strictly determine which items belong to which factors. Then, in order to assess the measurement model Convergent validity and Discriminant validity were conducted via confirmatory factor analyses. After the required prior analyses, Component based Structural Equation Modeling (Partial Least Square - PLS) was used to validate the predictive power of the proposed research model. Consequently, relationships between the influencing factors were detected and the results showed that the factors related with Belief dimension directly influenced behavioral intention of instructors. Also, the factors under the Individual, Social and Technological dimensions indirectly affected behavioral intention of users towards learning management system use. Additionally, structured and informal interviews were performed with ten instructors and the findings of the research model were explained with the opinions of system users. The indications of this research will be valuable for implementation, management and continuous improvement of learning management systems.

Ein zusammengesetzter Index ist eine aggregierte Variable, die aus individuellen Indikatoren und Gewichten besteht, wobei die Gewichte die relative Wichtigkeit jedes Indikators darstellen. Zusammengesetzte Indizes werden oft benutzt um latente Phänomene zu schreiben oder komplexe Informationen zu einer geringen Anzahl an Variablen zusammenzufassen. Es ist von großer Bedeutung richtige Gewichte für die Variablen, die einen zusammengesetzten Index bilden, zu wählen. Hauptkomponentenanalyse (PCA) ist ein populärer Ansatz um Gewichte abzuleiten, aber es ist ungeeignet, wenn informative Variationen nur kleine Varianzen der Variablen in einem zusammengesetzten Index haben. Deshalb schlägt diese Studie vor, Partial Least Squares (PLS) anzuwenden, welches die Beziehung zwischen Zielvariablen and den Variablen in einem zusammengesetzten Index ausnutzt. Unsere Simulationsstudie zeigt, dass PLS so gut wie PCA funktioniert oder erheblich es übertrifft. Zusätzlich sind in der Praxis die Variablen in einem zusammengesetzten Index häufig nicht-metrisch. Solche Variablen benötigen spezielle Verfahren, um PCA oder PLS anzuwenden. Diese Studie untersucht mehrere PCA und PLS Algorithmen für nicht-metrische Variablen in der vorliegenden Literatur und vergleicht sie durch umfangreiche Simulationsstudien, um Empfehlungen für die Praxis abzugeben. Dummy coding zeigt häufig zufriedenstellende Leistung im Vergleich zu komplizierteren Methoden. Als unsere Anwendungen betrachten wir Vermögen, Globalisierung, Geschlechtergleichheit und Korruption, indem PCA- und PLS-basierte zusammengesetzte Indizes angewendet werden. PLS erzeugt für die jeweiligen Zielvariablen massgeschnittene zusammengesetzte Indizes, die häufig bessere Leistung als PCA zeigten. Ein Vergleich zwischen PCA und PLS Gewichten und Koeffizienten zeigt, welche Variablen für die jeweiligen Zielvariablen besonders relevant sind.

The purpose of this analysis is to evaluate the usefulness of Fourier Transform Infrared (FTIR) spectroscopy in the detection of Herpes Simplex Virus 1 (hsv1) infection at an early stage. The raw absorption values were standardized to eliminate inter-sampling error. Wilcoxon-Mann-Whitney (WMW) statistic's Z score was calculated to select significant spectral regions. Partial least squares modeling was performed because of multicollinearity. Kolmogorov-Smirnov statistic showed models for healthy tissues from different time groups were not from same distribution. The additional 24 hour dataset was evaluated using the following methods. Variables were selected by WMW Z score. Difference of Composites statistic, DC, was created as a disease indicator and evaluated using area under the ROC curve, specificities, and confidence intervals using bootstrap algorithm. The specificity of DC was high, however the confidence intervals were large. Future studies are required with larger sample sizes to test this statistic's usefulness.