One of the challenges of data mining is finding hyper-parameters for a learning algorithm that will produce the best model for a given dataset. Hyper-parameter optimization automates this process, but it can still take significant time. It has been found that hyperparameter optimization does not always result in induced models with significant improvement over default hyper-parameters, yet no systematic analysis of the role of hyper-parameter optimization in machine learning has been conducted. We propose the use of meta-learning to inform the decision to optimize hyper-parameters based on whether default hyper-parameter performance can be surpassed in a given amount of time. We will build a base of metaknowledge, through a series of experiments, to build predictive models that will assist in the decision process.

Tese de mestrado integrado. Engenharia Electrotécnica e de Computadores (Energia). Universidade do Porto. Faculdade de Engenharia. 2011

Tese de mestrado integrado. Engenharia Electrotécnica e de Computadores. Faculdade de Engenharia. Universidade do Porto. 2009

Tese de mestrado integrado. Engenharia Electrotécnica e de Computadores (Major Energia). Faculdade de Engenharia. Universidade do Porto. 2009

Tese de doutoramento. Programa Doutoral em Engenharia Industrial e Gestão. Faculdade de Engenharia. Universidade do Porto. 2012

Tese de doutoramento. Engenharia Electrotécnica e de Computadores. Faculdade de Engenharia. Universidade do Porto. 2007

Alcohol misuse is a very common problem with high financial and personal costs. Treatment requires allocation of limited resources for optimal impact. Responsible decision making in this area should be based upon reasoned weighing of research evidence. Miller and colleagues completed a meta-analytic review of all controlled studies published before 1992 to help clinicians do just that. The coding system they employed examined methodological quality, as well as outcome, to obtain a rank­ordering of treatments that seem to have the most quality research support. The current study attempts to extend this work utilizing the same coding on studies published since 1992, and combine both databases of articles. Revised rank orderings of treatments and conclusions regarding variables related to outcomes are reported. Implications are discussed, along with limitations of this review. An upward trend in methodological quality over time was also discovered.

Often for prevention program designs, researchers are interested in understanding the processes through which a program impacts a targeted outcome. Mediation analysis assists in identifying not only how a program influences an outcome, but also which intermediate variables (i.e., mediators) cause the effects between a program and an outcome to occur. Mediation analysis explains why a program works, which is useful for program developers in creating effective prevention and intervention-based programs. To make use of mediation analysis findings for preventive intervention programs, researchers need a comprehensive understanding of the mediators between various programs and outcomes. However, a comprehensive examination into which mediators are most effective has yet to take place. This is likely due to the lack of theoretical and quantitative guidance on conducting a comprehensive comparison study for mediated effects. As such, this work establishes a framework for measuring mediated effects in a comprehensive context. This thesis establishes a framework under which to evaluate mediated effects across multiple studies, demonstrates the application of this framework, and discusses the broader implications of this approach. Identifying the most effective mediators through the proposed approach lends a valuable understanding to practitioners and policymakers about critical actions for preventing a given outcome.

The purpose of this study was to review the empirical research evidence on the effect of teacher certification on student achievement. An exploratory meta-analysis was conducted on studies that examined the effect of fully certified and less-than-fully certified teachers on student achievement. The meta-analysis focused on the areas of mathematics, science and reading and explored trends across areas of achievement, school level and research design. The study was directed towards (a) a synthesis of findings, and (b) recommendations for future research and policy decisions.The meta-analysis population consisted of five individual studies that generated twenty-seven effect size estimates. Three studies utilized either an individual level or class level of analysis and yielded twelve mean difference effect size estimates. Two studies utilized either a school or state level of analysis and yielded fifteen correlational effect size estimates. The majority of findings in mathematics favor the positive effect of fully certified teachers. In science, the findings pointed towards equivalent levels of student achievement for fully certified and less-than fully certified teachers. All the findings associated with reading favored the positive effect of fully certified teachers. It appears that certification may be more crucial to student achievement in reading and mathematics than in science. Across school levels, the overall trend suggests that full certification may be more crucial to student achievement in elementary school than middle or high school. Across levels of analysis and research design, studies that utilize an aggregate level of analysis yield a greater number of positive study outcomes than designs conducted at the individual or class level. A key finding is that given the specifications of the meta-analysis, direct evidence of the relationship between certification and student achievement is limited to five peer-reviewed, published studies. Additional findings illuminated several issues that are vital to improving the quantity and quality of research on teacher certification. Eight specific recommendations were directed towards academic researchers who plan to study the topic. Four recommendations are directed towards policy-makers at the state and federal level who are involved in setting standards and planning legislation for educator preparation.

Clearly, the wide range of health information sources on the World Wide Web has the potential to lead to distribution of inaccurate medical information from unqualified sources bringing a great risk. Given the growing number of Internet users that access health-related information, the need for a more standard means to validate web site content is apparent. This paper examines how source, information, timeliness, accessibility, and design factors impact web document credibility on a narrower health topic - Alternative Medicine. It also examines the contrasts of different levels of credibility with metadata usage as well as the relationships between metadata usage measures. These preliminary results and examinations give an overview of how metadata is currently being used in this subject area.