Dropout prediction in e-learning courses through the combination of machine learning techniques

In this paper, a dropout prediction method for e-learning courses, based on three popular machine learning techniques and detailed student data, is proposed. The machine learning techniques used are feed-forward neural networks, support vector machines and probabilistic ensemble simplified fuzzy ARTMAP. Since a single technique may fail to accurately classify some e-learning students, whereas another may succeed, three decision schemes, which combine in different ways the results of the three machine learning techniques, were also tested. The method was examined in terms of overall accuracy, sensitivity and precision and its results were found to be significantly better than those reported in relevant literature.     

An Efficient Newton-Krylov Implementation of the Constrained Runs Scheme for Initializing on a Slow Manifold

The long-term dynamic behavior of many dynamical systems evolves on a low-dimensional, attracting, invariant slow manifold, which can be parameterized by only a few variables (“observables”). The explicit derivation of such a slow manifold (and thus, the reduction of the long-term system dynamics) is often extremely difficult or practically impossible. For this class of problems, the equation-free framework has been developed to enable performing coarse-grained computations, based on short full model simulations. Each full model simulation should be initialized so that the full model state is consistent with the values of the observables and close to the slow manifold. To compute such an initial full model state, a class of constrained runs functional iterations was proposed (Gear and Kevrekidis, J. Sci. Comput. 25(1), 17–28, 2005; Gear et al., SIAM J. Appl. Dyn. Syst. 4(3), 711–732, 2005). The schemes in this class only use the full model simulator and converge, under certain conditions, to an approximation of the desired state on the slow manifold. In this article, we develop an implementation of the constrained runs scheme that is based on a (preconditioned) Newton-Krylov method rather than on a simple functional iteration. The functional iteration and the Newton-Krylov method are compared in detail using a lattice Boltzmann model for one-dimensional reaction-diffusion as the full model simulator. Depending on the parameters of the lattice Boltzmann model, the functional iteration may converge slowly or even diverge. We show that both issues are largely resolved by using the Newton-Krylov method, especially when a coarse grid correction preconditioner is incorporated.     

On the efficiency of user identification: a system-based approach

In the Internet era, users’ fundamental privacy and anonymity rights have received significant research and regulatory attention. This is not only a result of the exponential growth of data that users generate when accomplishing their daily task by means of computing devices with advanced capabilities, but also because of inherent data properties that allow them to be linked with a real or soft identity. Service providers exploit these facts for user monitoring and identification, albeit impacting users’ anonymity, based mainly on personal identifiable information or on sensors that generate unique data to provide personalized services. In this paper, we report on the feasibility of user identification using general system features like memory, CPU and network data, as provided by the underlying operating system. We provide a general framework based on supervised machine learning algorithms both for distinguishing users and informing them about their anonymity exposure. We conduct a series of experiments to collect trial datasets for users’ engagement on a shared computing platform. We evaluate various well-known classifiers in terms of their effectiveness in distinguishing users, and we perform a sensitivity analysis of their configuration setup to discover optimal settings under diverse conditions. Furthermore, we examine the bounds of sampling data to eliminate the chances of user identification and thus promote anonymity. Overall results show that under certain configurations users’ anonymity can be preserved, while in other cases users’ identification can be inferred with high accuracy, without relying on personal identifiable information.     

Effective-capacity-based stochastic delay guarantees for systems with time-varying servers, with an application to IEEE 802.11 WLANs

Many network applications rely on stochastic QoS guarantees. With respect to loss-related performance, the effective bandwidth/capacity theory has proved useful for calculating loss probabilities in queues with complex input and server processes and for formulating simple admission control tests to ensure associated QoS guarantees. This success has motivated the application of the theory for delay-related QoS too. However, up until now this application has been justified only heuristically for queues with variable service rate. The paper fills this gap by establishing rigorously that the effective bandwidth/capacity theory may be used for the asymptotically correct calculation and enforcement of delay tail-probabilities in systems with variable rate servers too. Subsequently, the paper applies the general results to IEEE 802.11 WLANs, by representing each IEEE 802.11 station as an On/Off server and employing the effective capacity function for this model. Comparison of analytical results with simulation validates the effectiveness of the On/Off IEEE 802.11 model for delay-related QoS, complementing earlier results on loss-related performance.     

GDF v2.0, an enhanced version of GDF

An improved version of the function estimation program GDF is presented. The main enhancements of the new version include: multi-output function estimation, capability of defining custom functions in the grammar and selection of the error function. The new version has been evaluated on a series of classification and regression datasets, that are widely used for the evaluation of such methods. It is compared to two known neural networks and outperforms them in 5 (out of 10) datasets.

Program summary

Title of program: GDF v2.0Catalogue identifier: ADXC_v2_0Program summary URL:http://cpc.cs.qub.ac.uk/summaries/ADXC_v2_0.htmlProgram obtainable from: CPC Program Library, Queen's University, Belfast, N. IrelandLicensing provisions: Standard CPC licence, http://cpc.cs.qub.ac.uk/licence/licence.htmlNo. of lines in distributed program, including test data, etc.: 98 147No. of bytes in distributed program, including test data, etc.: 2 040 684Distribution format: tar.gzProgramming language: GNU C++Computer: The program is designed to be portable in all systems running the GNU C++ compilerOperating system: Linux, Solaris, FreeBSDRAM: 200000 bytesClassification: 4.9Does the new version supersede the previous version?: YesNature of problem: The technique of function estimation tries to discover from a series of input data a functional form that best describes them. This can be performed with the use of parametric models, whose parameters can adapt according to the input data.Solution method: Functional forms are being created by genetic programming which are approximations for the symbolic regression problem.Reasons for new version: The GDF package was extended in order to be more flexible and user customizable than the old package. The user can extend the package by defining his own error functions and he can extend the grammar of the package by adding new functions to the function repertoire. Also, the new version can perform function estimation of multi-output functions and it can be used for classification problems.Summary of revisions: The following features have been added to the package GDF:

•

Multi-output function approximation. The package can now approximate any function . This feature gives also to the package the capability of performing classification and not only regression.

•

User defined function can be added to the repertoire of the grammar, extending the regression capabilities of the package. This feature is limited to 3 functions, but easily this number can be increased.

•

Capability of selecting the error function. The package offers now to the user apart from the mean square error other error functions such as: mean absolute square error, maximum square error. Also, user defined error functions can be added to the set of error functions.

•

More verbose output. The main program displays more information to the user as well as the default values for the parameters. Also, the package gives to the user the capability to define an output file, where the output of the gdf program for the testing set will be stored after the termination of the process.

Additional comments: A technical report describing the revisions, experiments and test runs is packaged with the source code.Running time: Depending on the train data.     

Three-dimensional face recognition in the presence of facial expressions: an annotated deformable model approach

In this paper, we present the computational tools and a hardware prototype for 3D face recognition. Full automation is provided through the use of advanced multistage alignment algorithms, resilience to facial expressions by employing a deformable model framework, and invariance to 3D capture devices through suitable preprocessing steps. In addition, scalability in both time and space is achieved by converting 3D facial scans into compact metadata. We present our results on the largest known, and now publicly available, face recognition grand challenge 3D facial database consisting of several thousand scans. To the best of our knowledge, this is the highest performance reported on the FRGC v2 database for the 3D modality