Applications of Topic Modelling Algorithms in Hybrid Machine Learning Frameworks to Forecast Customer Value in the Financial Industry
DR. MARCOS MACHADO
ASSISTANT PROFESSOR, IEBIS DEPARTMENT, UNIVERSITY OF TWENTE.
The increased volume of unstructured data, namely text, has led to the development of various techniques to extract valuable information from it. Specifically, Topic Modelling (TM) is one of the most popular branches of text analytics. In the financial sector, it has been applied to extract features from different sources (e.g., news, customer e-mails, and reviews). This information can be very useful in building predictive frameworks to assess Customer Relationship Management (CRM) metrics (e.g., credit risk, churn) in addition to the financial and behavioral variables commonly used in the industry. In this presentation, I will discuss the use of different TM algorithms (LDA, NMF, LSA, Top2Vec, and BERTopic) to extract valuable information from text-based features and analyze their impact on forecasting CRM metrics. The proposed frameworks use the extracted topics in individual and hybrid ML algorithms to predict customers' Risk-Adjusted Revenue (RAR). The individual models refer to various ML algorithms (e.g., Gradient Boosting, Adaboost, etc.) used to forecast RAR; alternatively, the hybrid frameworks contain clustering methods implemented before predicting RAR with the same set of individual ML algorithms. Results show that hybrid ML frameworks can outperform individual ML methods in predictive power and provide managers with many customer portfolios with different levels of risk and return. In particular, the combination of text and hard features boosts predictions. Also, the ML models we implement are high predictors and achieve an R2 over 85% for all cases. Finally, all extended RAR models achieve better predictive power than does the baseline model used in the literature.
Marcos Machado is an Assistant Professor in Business and Information Systems at the Industrial Engineering and Business Information Systems (IEBIS) department section of the University of Twente. He holds a Ph.D. in Modelling and Computational Science from the Ontario Tech University (Canada), a MSc in Production Engineering from the University of Sao Paulo (Brazil), and a BSc in Mathematics from the Federal Institute of Education, Science, and Technology of Ceara (Brazil). He also has over seven years of experience working in the Brazilian and Canadian banking industries. His main research interests are focused on applications of Artificial Intelligence (AI) and analytics to solve business problems.
A Generic Optimization Model for Operational Planning and Bidding of Production Units and Evaluation of Flexibility in District Heating Systems
DR. DANIELA GUERICKE
ASSISTANT PROFESSOR, IEBIS DEPARTMENT, UNIVERSITY OF TWENTE.
District heating is an important component in the EU strategy to reach the set emission goals, since it allows an efficient supply of heat while using the advantages of sector coupling between different energy carriers such as power, heat, gas and biomass. Most district heating systems use several different types of units to produce heat. The technologies reach from natural gas and electric boilers to biomass-fired units as well as waste heat from industrial processes and solar thermal units. Furthermore, combined heat and power units (CHP) units are often included to exploit synergy effects of simultaneous heat and electricity production. Daniela will present a generic network-flow based mathematical formulation for the operational production optimization in district heating systems. The generality of the model allows it to be used for most district heating systems although they might use different combinations of technologies in different system layouts. The mathematical formulation is based on stochastic programming to account for the uncertainty of production from non-dispatchable units such as solar thermal units. Daniela will show how the model can be used for determining bids for electricity markets and evaluation of demand-side flexibility in district heating systems. The results are based on real data from district heating systems in Denmark with different requirements, which illustrates an application under real-world system configurations.
Daniela Guericke is Assistant Professor for Stochastic Operations Research at the Section Industrial Engineering and Business Information Systems, University of Twente. Her research focuses on (stochastic) operations research and optimization in application areas such as energy systems and health care. In particular, she is interested in decision-making under uncertainty and solving large-scale optimization problems. Daniela received her PhD in Business Information Systems from the Decision Support and Operations Research Lab, Paderborn University. Afterwards, she worked as a postdoctoral researcher at the Department of Applied Mathematics and Computer Science, Technical University of Denmark (DTU). In 2020, she became Assistant Professor for Decision-making under Uncertainty in Integrated Energy Systems at DTU. In 2021, Daniela received the Young Researchers Award of the German OR Society (GOR e.V.)