I put a focus on data mining in the marine domain. This line of research is funded by the SAMS PhD program and is a cooperation with the MSYS group from ICBM, University of Oldenburg. The MSYS group provides large data sets, e.g., sensor data from the Time Series Station near Spiekeroog, which suffer from missing data and noise. I repair this data by combining classical interpolation methods with machine learning approaches. My algorithms learn about trends in data by utilizing discretization. So far, I successfully repaired multivariate time series data with consecutively missing values and detected events of research interest for MSYS. Next, I aim to replace temporary missing or failing sensors by our sensor predictions.
Stefan Oehmcke, M.Sc.
machine learning for marine time series:
- imputation of consecutively missing values
- preprocessing of time series via time dimensionality reduction
- prediction of marine sensor values with time-spatial distance
general deep learning:
- recurrent neural networks
- convolutional nets
- Stefan Oehmcke, Oliver Zielinski, Oliver Kramer: "kNN ensembles with penalized DTW for multivariate time series imputation", International Joint Conference on Neural Networks (IJCNN), 2774-2781, 2016, IEEE
- Stefan Oehmcke, Justin Heinermann, Oliver Kramer: "Analysis of Diversity Methods for Evolutionary Multi-objective Ensemble Classifiers", Applications of Evolutionary Computation, 567-578, 2015, Springer
- Stefan Oehmcke, Oliver Zielinski, Oliver Kramer: "Event Detection in Marine Time Series Data", KI 2015: Advances in Artificial Intelligence, 279-286, 2015, Springer
- Seminar: Data Mining For Maritime Applications