Gerd Kortemeyer

Predicting student performance: an application of data mining methods with an educational Web-based system

Newly developed Web-based educational technologies offer researchers unique opportunities to study how students learn and what approaches to learning lead to success. Web-based systems routinely collect vast quantities of data on user patterns, and data mining methods can be applied to these databases. This paper presents an approach to classifying students in order to predict their final grade based on features extracted from logged data in an education Web-based system. We design, implement, and evaluate a series of pattern classifiers and compare their performance on an online course dataset. A combination of multiple classifiers leads to a significant improvement in classification performance. Furthermore, by learning an appropriate weighting of the features used via a genetic algorithm (GA), we further improve prediction accuracy. The GA is demonstrated to successfully improve the accuracy of combined classifier performance, about 10 to 12% when comparing to non-GA classifier. This method may be of considerable usefulness in identifying students at risk early, especially in very large classes, and allow the instructor to provide appropriate advising in a timely manner.

Experiences using the open-source learning content management and assessment system LON-CAPA in introductory physics courses

We discuss the development and functionality of the LON-CAPA system with a particular focus on its homework and examination functionality. We also describe its more general approach to course management and its infrastructure for course content sharing and reuse. We then focus on measures of student learning and the effectiveness of different content types.

Optimizing Classification Ensembles via a Genetic Algorithm for a Web-Based Educational System

Classification fusion combines multiple classifications of data into a single classification solution of greater accuracy. Feature extraction aims to reduce the computational cost of feature measurement, increase classifier efficiency, and allow greater classification accuracy based on the process of deriving new features from the original features. This paper represents an approach for classifying students in order to predict their final grades based on features extracted from logged data in an educational web-based system. A combination of multiple classifiers leads to a significant improvement in classification performance. By weighing feature vectors representing feature importance using a Genetic Algorithm (GA) we can optimize the prediction accuracy and obtain a marked improvement over raw classification. We further show that when the number of features is few, feature weighting and transformation into a new space works efficiently compared to the feature subset selection. This approach is easily adaptable to different types of courses, different population sizes, and allows for different features to be analyzed.

An analysis of asynchronous online homework discussions in introductory physics courses

Asynchronous online student discussions of online homework problems in introductory physics courses are analyzed with respect to course type, student course performance, student gender, problem difficulty, and problem type. It is found that these variables can significantly change the character of online student collaborations.

The Challenge of Teaching Introductory Physics to Premedical Students

Most physics instructors are motivated by a genuine interest in their subject area and in using physics to understand real-world phenomena. While many premedical students may share these interests, most are motivated by fulfilling their degree requirements and gaining admittance into medical school. To achieve this latter goal, they need excellent grades and have to do well on the Medical College Admissions Test (MCAT),1 which includes a physical sciences section that may not coincide with the learning goals of many physics courses. Too often both sides simply give up, and courses come to some kind of unspoken agreement of how to go through the motions of completing the course with the least amount of mutual aggravation, while real physics falls by the wayside. But how exactly does this discrepancy manifest itself, and what—if anything other than giving up—can be done about it? In this paper, I first survey learner beliefs, expectations, and preferences and then attempt to identify approaches and resources that may partly address the identified issues.

Nuclear flow in consistent Boltzmann algorithm models

We investigate the stochastic Direct Simulation Monte Carlo method (DSMC) for numerically solving the collision-term in heavy-ion transport theories of the Boltzmann-Uehling-Uhlenbeck (BUU) type. The first major modification we consider is changes in the collision rates due to excluded volume and shadowing/screening effects (Enskog theory). The second effect studied by us is the inclusion of an additional advection term. These modifications ensure a non-vanishing second virial and change the equation of state for the scattering process from that of an ideal gas to that of a hard-sphere gas. We analyse the effect of these modifications on the calculated value of directed nuclear collective flow in heavy ion collisions, and find that the flow slightly increases.

Revisiting the Geoscience Concept Inventory: A call to the community

The use of concept inventories in science and engineering has fundamentally changed the nature of instructional assessment. Nearly a decade ago, we set out to establish a baseline for widespread and integrated assessment of entry-level geoscience courses. The result was the first Geoscience Concept Inventory (GCI v.1.0). We are now retiring GCI v.1.0 and rebuilding the GCI as a more community-based, comprehensive, and effective instrument. We are doing this in the hopes that GCI users, many of whom have expressed a need for a revised and expanded instrument, and the geoscience community at large will view it as a springboard for collaborative action and engagement. If we work together as collaborators, the geosciences have the potential to evaluate learning across our community and over time. INTRODUCTION The Geoscience Concept Inventory (GCI; Fig. 1) was developed to diagnose conceptual understanding and assess learning in entry-level geoscience courses. The GCI has become a staple in many classroom-based research studies, is being revised for use in pre-college settings, and has been shown to discriminate between experts and novices. Although a valuable research tool, the GCI is in need of an expansion that can only be accomplished by a community of geoscientists and educators working together. This paper is a call for that collaboration. The GCI holds a unique place in the concept inventory world for several reasons. First, the GCI is the only concept inventory to generate a bank of correlated concept inventory questions for higher education science (Libarkin and Anderson, 2006). Through this correlation, users of the GCI can create course-specific subtests rather than being tied to a single set of questions. Second, the GCI contains single response, two-tier, and multiple-response multiple-choice questions. Two-tier questions offer added insight into student thinking by requesting an explanation for student responses (Treagust, 1988). Multipleresponse questions, essentially a set of true/false items, are generally more difficult than typical single-response items and are cognitively similar to free response questions, offering deeper insight into cognition (Kubinger and Gottschall, 2007). Third, GCI questions were developed from ideas that both experts and novices found important for entry-level geoscience courses. A review of textbooks provided initial ideas about important concepts for inclusion on the GCI, while open-ended interviews with students provided additional topics (Libarkin and Anderson, 2005). For example, in-depth interviews suggest that students conflate gravity and magnetism and inflate the importance of magnetic fields on the movement of large objects. Addressing this mixing and mis-scaling is important for student understanding of geomagnetism and its effects, a discovery that only became apparent after considering the student perspective. THE NEED TO REVISE AND EXPAND THE GCI AS A COMMUNITY The original GCI questions were piloted with up to 5,000 students enrolled at >40 institutions nationwide, with the current version in use by >200 faculty and researchers. The GCI has been used to estimate learning in geoscience courses, including evaluation of specific instructional approaches (e.g., Kortz et al., 2008) and analysis of learning (e.g., Petcovic and Ruhf, 2008). In ongoing work, GCI scores have been shown to correlate strongly with geological mapping ability. This suggests that the GCI, a measure of very foundational knowledge, can be used as a skills measure to predict performance on an expert task. While we are encouraged that GCI v.1.0 was useful in these studies, we acknowledge that the instrument ingrains our own biases and limitations. As many of our colleagues have stated, the GCI is both an effective instrument for gauging learning in entry-level geoscience courses and a test in need of revision. The diversity of geoscience courses at all levels should be reflected in the assessment instruments used to evaluate learning nationwide. Expansion to more complex, wider ranging questions will allow replicable assessment in advanced courses and across geoscience programs. A critical need for questions targeted toward upper-level courses requires community GSA Today, v. 21, no. 8, doi: 10.1130/G110GW.1 *libarkin@msu.edu Revisiting the Geoscience Concept Inventory: A call to the community GROUNDWORK T H E G EO LO GI CAL SCIETY OF AM ERIC A Furthering the Inf luence of Earth Science

The LearningOnline Network with CAPA Initiative

With funding of the National Science Foundation Information Technology Research program, the LearningOnline Network with CAPA (LON-CAPA) Initiative aims to create and sustain a cross-institutional distributed platform for content creation, sharing, and delivery. LON-CAPA will allow groups of organizations (schools, departments, universities, commercial business) to link their online instructional resources in a common marketplace, thus creating an online economy for instructional resources.

Causality violations in cascade models of nuclear collisions

Transport models have successfully described many aspects of intermediate energy heavy-ion collision dynamics. As the energies increase in these models to the ultrarelativistic regime, Lorentz covariance and causality are not strictly respected. The standard argument is that such effects are not important to final results; but they have not been seriously considered at high energies. We point out how and why these happen, how serious of a problem they may be and suggest ways of reducing or eliminating the undesirable effects.

Virial Corrections to Simulations of Heavy Ion Reactions

Within quantum molecular dynamics (QMD) simulations we demonstrate the effect of virial corrections on heavy ion reactions. Unlike in standard codes, the binary collisions are treated as nonlocal so that the contribution of the collision flux to the reaction dynamics is covered. A comparison with standard QMD simulations shows that the virial corrections lead to a broader proton distribution bringing theoretical spectra closer towards experimental values. Complementary Boltzmann-Uehling-Uhlenbeck simulations reveal that the nonlocality enhances the collision rate in the early stage of the reaction. It suggests that the broader distribution appears due to an enhanced preequilibrium emission of particles. [S0031-9007(99)09104-8] The Boltzmann equation including the Pauli blocking [the Boltzmann-Uehling-Uhlenbeck (BUU) equation [1]] and the closely related method of quantum molecular dynamics (QMD) [2,3] are extensively used to interpret experimental data from heavy ion reactions. Because of their quasiclassical character, they offer a transparent picture of the internal dynamics of reactions and allow one to link observed particle spectra with individual stages of reactions. The expectation to cover the heavy ion reactions within experimental errors has been recently set back by a failure of BUU simulations to describe the energy and angular distribution of neutrons and protons in the energy domain #200 MeVyA [4‐6]. Indeed, the Boltzmann equation does not contain all of the relevant physics. As noticed in numerical studies of hard sphere cascades by Halbert [7] and, more generally, by Malfliet [8], it is unfortunate that all dynamical models rely more or less on the use of the space- and time-local approximation of binary collisions inherited from the Boltzmann equation. This approximation neglects a contribution of the collision flux to the compressibility and the shear viscosity which control the hydrodynamic motion during the reaction. In order to include the collision flux and other virial corrections, the nonlocal character of binary collisions has to be accounted for. Malfliet also demonstrated that nonlocal collisions can be easily incorporated into BUU simulation codes.