Forschungsgebiete:

• Machine Learning, Deep Learning und künstliche Intelligenz, insb. Hyperparametertuning & Interpretierbarkeit von Black Box Modellen
• Computational Statistics
• Causal Inference
• Text Mining & Sentimentanalyse

Angewandte Forschung:

• Credit Scoring & IRBA
• Customer Relationship Management
• Industrie 4.0
• Medizinische Diagnostik
• Music Information Retrieval
• Spracherkennung

In the recent years several free open source tools for credit scorecard modelling have become available. An overview on the landscape of R packages for this purpose can be found here. On the other hand modern machine learning methods have entered the application field of credit scoring. A comprehensive study with focus on the imbalance of the target class and hyperparameter tuning can be found under this link. From a regulatory perspective, transparency of risk models is required. A corresponding framerwork has presented here while possibilities to build fair credit scores are investigated here.

There is an ongoing hype around artificial intelligence and machine learning and open source autoML tools have made machine learning easily accessible to anyone. Nevertheless it has to be emphasized that a profound understanding of the models should be part of each model development and requires for experienced analysts. In general, understanding of models as they result from many recently proposed machine learning algorithms becomes difficult as the model typically come along as black boxes. I gave a short general overview on the topic at the Predictive Analytics World Conference on 18./19.11.19 in Berlin. Two quite general frameworks that might help understanding can be found here and here.

• A general framework for the use of XAI in order to improve transparency and auditability of ML models in credit scoring is available here.
• Interpretable Machine Learning has also been the focus theme of the AG DANK 2018 fall symposium (cf. below). The results of its satellite contest can be found here.
• Although interpretability is important it should be done carefully as it is outlined in a paper on quantifying explainability of a ML model by partial dependence plots. A corresponding R package can be downloaded here.
• An analysis of fairness correction for credit risk sorring is availble under this link.

For a list of my publications cf.e.g.Google Scholar or  ResearchGate.

Teilprojekt im Rahmen des Landesprogramms Digitale Lehre in Mecklenburg/Vorpommern (Laufzeit: 1.3.2020 bis 31.12.2023).

Untersuchungsgegenstand des Projektes sind touristische Entwicklungsprojekte in Transformations- und Schwellenländern. Mit Hilfe von Instrumentarien des Text Mining sollen  wahrgenommene Qualitätskriterien der Nutzer identifiziert und untersucht werden. Kooperation mit der HTW Berlin. Projektlaufzeit: 1.8.2019 - 31.7.2020.

Paper

Umsetzung des WWM Algorithmus zur Identifikation einer Wechselwirkungsmatrix von Herb-drug interaktionen im Rahmen eines Unterauftrags des Kompetenznetzes Komplementärmedizin in der Onkologie (KOKON). Projektlaufzeit: 01.07.2019 - 31.08.2019.

• HanseYachts AG

• Mecklenburger Backstuben

• Strelapark

• Tourismuszentrale Rostock & Warnemünde

• Vogelpark Marlow

• Waren (Müritz) Kur- und Tourismus GmbH