Andre Busche

Improving Academic Performance Prediction by Dealing with Class Imbalance

This paper introduces and compares some techniques used to predict the student performance at the university. Recently, researchers have focused on applying machine learning in higher education to support both the students and the instructors getting better in their performances. Some previous papers have introduced this problem but the prediction results were unsatisfactory because of the class imbalance problem, which causes the degradation of the classifiers. The purpose of this paper is to tackle the class imbalance for improving the prediction/classification results by over-sampling techniques as well as using cost-sensitive learning (CSL). The paper shows that the results have been improved when comparing with only using baseline classifiers such as Decision Tree (DT), Bayesian Networks (BN), and Support Vector Machines (SVM) to the original datasets.

Buried pipe localization using an iterative geometric clustering on GPR data

Ground penetrating radar is a non-destructive method to scan the shallow subsurface for detecting buried objects like pipes, cables, ducts and sewers. Such buried objects cause hyperbola shaped reflections in the radargram images achieved by GPR. Originally, those radargram images were interpreted manually by human experts in an expensive and time consuming process. For an acceleration of this process an automatization of the radargram interpretation is desirable. In this paper an efficient approach for hyperbola recognition and pipe localization in radargrams is presented. The core of our approach is an iterative directed shape-based clustering algorithm combined with a sweep line algorithm using geometrical background knowledge. Different to recent state of the art methods, our algorithm is able to ignore background noise and to recognize multiple intersecting or nearby hyperbolas in radargram images without prior knowledge about the number of hyperbolas or buried pipes. The whole approach is able to deliver pipe position estimates with an error of only a few millimeters, as shown in the experiments with two different data sets.

GamRec: A Clustering Method Using Geometrical Background Knowledge for GPR Data Preprocessing

GPR is a nondestructive method to scan the subsurface. On the resulting radargrams, originally interpreted manually in a time consuming process, one can see hyperbolas corresponding to buried objects. For accelerating the interpretation a machine shall be enabled to recognize hyperbolas on radargrams autonomously. One possibility is the combination of clustering with an expectation maximization algorithm. However, there is no suitable clustering algorithm for hyperbola recognition. Hence, we propose a clustering method specialized for this problem. Our approach is a directed shape based clustering combined with a sweep line algorithm. In contrast to other approaches our algorithm finds hyperbola shaped clusters and is (1) able to recognize intersecting hyperbolas, (2) noise robust and (3) does not require to know the number of clusters in the beginning but it finds this number. This is an important step towards the goal to fully automatize the buried object detection.

Active learning for technology enhanced learning

Suggesting tasks and learning resources of appropriate difficulty to learners is challenging. Neither should they be too difficult and nor too easy. Well-chosen tasks would enable a quick assessment of the learner, well-chosen learning resources would speed up the learning curve most. We connect active learning to classical pedagogical theory and propose the uncertainty sampling framework as a means to the challenge of selecting optimal tasks and learning resources to learners. To assess the efficiency of this strategy, we compared different exercise selection strategies and evaluated their effect on different datasets. We consistently find that uncertainty sampling significantly outperforms several alternative exercise selection approaches and thus leads to a faster convergence to the true assessment. These findings demonstrate that active (machine) learning is consistent with classic learning theory. It is a valuable instrument for choosing appropriate exercises as well as learning resources both from a teachers and from a learners perspective.

Calculation of upper esophageal sphincter restitution time from high resolution manometry data using machine learning

OBJECTIVE After swallowing, the upper esophageal sphincter (UES) needs a certain amount of time to return from maximum pressure to the resting condition. Disturbances of sphincter function not only during the swallowing process but also in this phase of pressure restitution may lead to globus sensation or dysphagia. Since UES pressures do not decrease in a linear or asymptotic manner, it is difficult to determine the exact time when the resting pressure is reached, even when using high resolution manometry (HRM). To overcome this problem a Machine Learning model was established to objectively determine the UES restitution time (RT) and moreover to collect physiological data on sphincter function after swallowing. METHODS AND MATERIAL HRM-data of 15 healthy participants performing 10 swallows each were included. After manual annotation of the RT interval by two swallowing experts, data were transferred to the Machine Learning model, which applied a sequence labeling modeling approach based on logistic regression to learn and objectivize the characteristics of all swallows. Individually computed RT values were then compared with the annotated values. RESULTS Estimates of the RT were generated by the Machine Learning model for all 150 swallows. When annotated by swallowing experts mean RT of 11.16s±5.7 (SD) and 10.04s±5.74 were determined respectively, compared to model-generated values from 8.91s±3.71 to 10.87s±4.68 depending on model selection. The correlation score for the annotated RT of both examiners was 0.76 and 0.63 to 0.68 for comparison of model predicted values. CONCLUSIONS Restitution time represents an important physiologic swallowing parameter not previously considered in HRM-studies of the UES, especially since disturbances of UES restitution may increase the risk of aspiration. The data presented here show that it takes approximately 9 to 11s for the UES to come to rest after swallowing. Based on maximal RT values, we demonstrate that an interval of 25-30s in between swallows is necessary until the next swallow is initiated. This should be considered in any further HRM-studies designed to evaluate the characteristics of individual swallows. The calculation model enables a quick and reproducible determination of the time it takes for the UES to come to rest after swallowing (RT). The results of the calculation are partially independent of the input of the investigator. Adding more swallows and integrating additional parameters will improve the Machine Leaning model in the future. By applying similar models to other swallowing parameters of the pharynx and UES, such as the relaxation time of the UES or the activity time during swallowing, a complete automatic evaluation of HRM-data of a swallow should be possible.

A Unifying Framework for GPR Image Reconstruction

Ground Penetrating Radar (GPR) is a widely used technique for detecting buried objects in subsoil. Exact localization of buried objects is required, e.g. during environmental reconstruction works to both accelerate the overall process and to reduce overall costs. Radar measurements are usually visualized as images, so-called radargrams, that contain certain geometric shapes to be identified.This paper introduces a component-based image reconstruction framework to recognize overlapping shapes spanning over a convex set of pixels. We assume some image to be generated by interaction of several base component models, e.g., hand-made components or numerical simulations, distorted by multiple different noise components, each representing different physical interaction effects.We present initial experimental results on simulated and real-world GPR data representing a first step towards a pluggable image reconstruction framework.

Event Prediction in Pharyngeal High-Resolution Manometry

A prolonged phase of increased pressure in the upper esophageal sphincter (UES) after swallowing might result in globus sensation. Therefore, it is important to evaluate restitution times of the UES in order to distinguish physiologic from impaired swallow associated activities. Estimating the event \(t^{\star }\) where the UES has returned to its resting pressure after swallowing can be accomplished by predicting if swallowing activities are present or not. While the problem, whether a certain swallow is pathologic or not, is approached in Mielens (J Speech Lang Hear Res 55:892–902, 2012), the analysis conducted in this paper advances the understanding of normal pharyngoesophageal activities.

Hough Transform and Kirchhoff Migration for Supervised GPR Data Analysis

Ground penetrating radar (GPR) is a widely used technology for detecting buried objects in the subsoil. Radar measurements are usually depicted as radargram images, including distorted hyperbola-like shapes representing pipes running non-parallel to the measurement trace. Also because of the heterogeneity of subsoil, human experts are usually analysing radargrams only in a semi-automatic way by adjusting parameters of the detection models (exposed by the software used) to get best detection results. To gain a set of approximate hyperbola apex positions, unsupervised methods such as the Hough transform (HT) or Kirchhoff Migration are often used. By having high-quality, large-scale real-world measurement data collected on a specialized test site at hand, we both (a) analyse differences and similarities of the HT and Kirchhoff Migration quantitatively and analytically with respect to different preprocessing techniques, and (b) embed results from either technique into a supervised framework. The primary contribution of this paper is the conduction of an exhaustive experiment, not only showing their equivalence, but also showing that their application for the automated analysis of GPR data, unlike it is currently assumes, does not improve the detection performance significantly.