Search

Abstract The tremendous growth in electronic educational data creates the need to have meaningful information extracted from it. Educational Data Mining (EDM) is an exciting research area that can reveal valuable knowledge from educational databases. This knowledge can be used for many purposes, including identifying dropouts or weak students who need special attention and discovering extraordinary students who can be offered lifetime opportunities. Although former studies in EDM used an extensive range of features for predicting students’ academic achievement (in terms of (i) achieved grades or (ii) passing and failing), those features are sometimes not obtainable for practical usage, and therefore, the prediction models are not feasible for employment. This study uses data mining (DM) algorithms to predict the academic performance of master’ s students by using a non-extensive data set and including only the features that are easy to collect at the beginning of a studying program. To perform this study, we have collected over 700 students' records from 2010 to 2018 from the Faculty of Business Informatics and Mathematics at the University of Mannheim in Germany. Those records include demographics and post-enrollment features such as semester grades. The empirical results show the following: (i) the most significant features for predicting students' academic achievements are the students’ grades in each semester (importance rate between 14 and 36%), followed by the distance from students’ accommodation to university (importance rate between 6 and 18%) and culture (importance rate between 7 and 17%). On the other hand, gender, age, the numbers of failed courses, and the number of registered and unregistered exams per semester are less significant for the predictions. (ii) As expected, predictions performed after the second semester is more accurate than those performed after the first semester. (iii) Unsurprisingly, models that predict two classes yield better results than those that predict three. (iv) Random Forest classifier performs the best in all prediction models (0.77–0.94 accuracy), and using oversampling methods to deal with imbalanced data can significantly improve the performance of DM methods. For future work, we recommend testing the predictive models on other master programs and a larger datasets. Furthermore, we recommend investigating other oversampling approaches.

Advances in wearable technologies have the ability to revolutionize and improve people’s lives. The gains go beyond the personal sphere, encompassing business and, by extension, the global economy. The technologies are incorporated in electronic devices that collect data from consumers’ bodies and their immediate environment. Human activities recognition, which involves the use of various body sensors and modalities either separately or simultaneously, is one of the most important areas of wearable technology development. In real-life scenarios, the number of sensors deployed is dictated by practical and financial considerations. In the research for this article, we reviewed our earlier efforts and have accordingly reduced the number of required sensors, limiting ourselves to first-person vision data for activities recognition. Nonetheless, our results beat state of the art by more than 4% of F1 score.

The ongoing electrification of mobility comes with the challenge of charging electric vehicles (EVs) sufficiently while charging infrastructure capacities are limited. Smart charging algorithms produce charge plans for individual EVs and aim to assign charging capacities fairly and efficiently between vehicles in a fleet. In practice, EV charging processes follow nonlinear charge profiles such as constant-current, constant-voltage (CCCV). Smart charging must consider charge profiles in order to avoid gaps between charge plans and actual EV power consumption. Generally valid models of charge profiles and their parameters for a diverse set of EVs are not publicly available. In this work we propose a data-driven approach for integrating a machine learning model to predict arbitrary charge profiles into a smart charging algorithm. We train machine learning models with a dataset consisting of charging processes from the workplace gathered in 2016–2018 from a heterogeneous EV fleet of 1001 EVs with 18 unique models. Each charging process includes the time series of charging power. After preprocessing, the dataset contains 10.595 charging processes leading to 1.2 million data points in total. We then compare different machine learning models for charge profile predictions finding that XGBoost yields the most accurate predictions with a mean absolute error (MAE) of 126W and a relative MAE of 0.06. Simulations show that smart charging with the integrated XGBoost model leads to a more effective infrastructure usage with up to 21% more energy charged compared to smart charging without considering charge profiles. Furthermore, an ablation study on regression model features shows the EV’s model is not a necessary attribute for accurate charge profile predictions. However, charging features are required including the number of phases used for charging.

In the field of pervasive computing, wearable devices have been widely used for recognizing human activities. One important area in this research is the recognition of activities of daily living where especially inertial sensors and interaction sensors (like RFID tags with scanners) are popular choices as data sources. Using interaction sensors, however, has one drawback: they may not differentiate between proper interaction and simple touching of an object. A positive signal from an interaction sensor is not necessarily caused by a performed activity e.g., when an object is only touched but no interaction occurred afterwards. There are, however, many scenarios like medicine intake that rely heavily on correctly recognized activities. In our work, we aim to address this limitation and present a multimodal egocentric-based activity recognition approach. Our solution relies on object detection that recognizes activity-critical objects in a frame. As it is infeasible to always expect a high quality camera view, we enrich the vision features with inertial sensor data that monitors the users’ arm movement. This way we try to overcome the drawbacks of each respective sensor. We present our results of combining inertial and video features to recognize human activities on different types of scenarios where we achieve an F 1 -measure of up to 79.6%.

Working with multimodal datasets is a challenging task as it requires annotations which often are time consuming and difficult to acquire. This includes in particular video recordings which often need to be watched as a whole before they can be labeled. Additionally, other modalities like acceleration data are often recorded alongside a video. For that purpose, we created an annotation tool that enables to annotate datasets of video and inertial sensor data. In contrast to most existing approaches, we focus on semi-supervised labeling support to infer labels for the whole dataset. This means, after labeling a small set of instances our system is able to provide labeling recommendations. We aim to rely on the acceleration data of a wrist-worn sensor to support the labeling of a video recording. For that purpose, we apply template matching to identify time intervals of certain activities. We test our approach on three datasets, one containing warehouse picking activities, one consisting of activities of daily living and one about meal preparations. Our results show that the presented method is able to give hints to annotators about possible label candidates.