NSBE - Business Analytics
URI permanente para esta coleção:
Navegar
Entradas recentes
- Adversarial generative forecasting of daily Fraud Amount for sparse transaction time seriesPublication . Mueller, Moritz; Xufre, PatríciaIn collaboration with SIBS, this project forecasts daily accepted fraud amounts in e-commerce transactions to support proactive risk management. Utilizing a dataset ofover 166 million transactions (2023–2024), we engineered behavioral features to benchmark multiple machine learning models. XGBoost was the champion model, achievinga 16.56% MAPE, but struggling during volatility spikes. To improve reliability duringspike periods, we investigated three complementary strategies: GAN-based data augmentation to increase exposure to synthetic high-fraud scenarios, quantile-based forecasting to model the upper tail of the distribution, and transfer learning approaches that adapt large pre-trained time-series models to our use case.
- Using machine learning to Solve Real Banking challenges at Banco PrimusPublication . Rudolf, Nils; Batikas, MichailThis thesis examines how machine-learning models and explainable AI can be used to analyze two distinct use cases: loan conversion and cross-selling in retail banking. Using proprietary data from Banco Primus, logistic regression, random forest, and XGBoost models are evaluated using business-oriented back-testing. SHAP is applied to explain predictions and identify key drivers. Approved loan conversion is mainly associated with partner characteristics, process timing, and communication availability. Personal loan cross-selling is mainly associated with external credit profiles and behavioral history, revealing campaign fatigue. The findings support process optimization in loan origination and propensity-based targeting frameworks for cross-selling.
- Predicting approved loan conversionPublication . Pachelbel, Valeska Von; Batikas, MichailThis thesis examines how machine-learning models and explainable AI can be used to analyze two distinct use cases: loan conversion and cross-selling in retail banking. Using proprietary data from Banco Primus, logistic regression, random forest, and XGBoost models are evaluated using business-oriented back-testing. SHAP is applied to explain predictions and identify key drivers. Approved loan conversion is mainly associated with partner characteristics, process timing, and communication availability. Personal loan cross-selling is mainly associated with external credit profiles and behavioral history, revealing campaign fatigue. The findings support process optimization in loan origination and propensity-based targeting frameworks for cross-selling.
- Political polarization in Portugal-measuring affective polarization using large language modelsPublication . Gerth, Simon; Shen, YufeiPolitical polarization has become a growing concern in Portugal as new parties reshape parliamentary competition. This thesis examines how affective polarization in the Portuguese parliament evolves over time and how it aligns with the rise and consolidation of the radical right party Chega. Using the ParlaMint-PT corpus (2015–2024), outgroup-directed hostility in parliamentary interventions is measured with a calibrated ensemble combining a fine-tuned XLM-RoBERTa classifier and an instruction-based GPT-5-Nano model. The findings indicate stable levels before 2019, a decline around 2020–2021, and a sharp increase from 2022 onward, consistent with a delayed association with Chega’s consolidation.
- Analyzing rhetorical framing in brexit debates in the house of commons of the United Kigdom iusing large models: temporal trends and structural patterns-empirical insights from automated frame detectionPublication . Speckmann, Lina; Shen, YufeiThis study examines rhetorical framing in House of Commons debates on the United Kingdom's exit (Brexit) from the European Union from 2012 to 2022, using Laqrge Language Models(LLMs) to analyze 62,847 parliamentary speech chunks. Building on Neuman et al.'s (1992) generic frame typology, it investigates how framing strategies have changed over time and varied across parties, referendum positions, and speaker attributes. The computational approach provides scale and precision beyond traditional qualitative methods, showing that Economic and Conflict frames dominated and that shifts in usage tracked key Brexit milestones. The findings demonstrate LLMs' value for large-scale political discourse analysis and clarify how parliamentary actors constructed Brexit's meaning during a period of constitutional turmoil.
- Data excellence in operations management: enabling data-driven decision-making by developing a leadership dashboard at quantum systemsPublication . Stark, Lukas Tobias; Obermeier, Daniel; Rietenbach, Alexandra M.E.This work project designs and evaluates a leadership dashboard to support data-driven decision making in Quantum Systems’ Operations division. Guided by the research question “How can a leadership dashboard be designed to improve data insights in Operations management?”, the project follows a three-step approach inspired by Design Science Research. It derives a KPI framework, assesses data readiness across core systems, and implements selected indicators in Power BI. Evaluation along four data pillars: Volume, Quality, Accessibility, and Governance, combines quantitative comparisons and stakeholder feedback, indicating reduced information overload and fragmentation, increased data trust and improved accessibility of operational information for leadership.
- Seasonal momentum in equity markets: the role of calendar, earnings and monetary policy effectsPublication . Walch, David; Januário, AfonsoThis study examines whether combining distinct seasonality drivers across macroeconomic factors such as monetary policy, microeconomic factors such as earnings announcement intensity, and historical equity return patterns can improve systematic trading performance. While the combined signal historically delivered stronger risk-adjusted returns than its individual components and a buy-and-hold benchmark, its effectiveness declines sharply in recent decades. The results suggest that those market anomalies have weakened over time, consistent with rising market efficiency. These results are in line with research showing that that return anomalies tend to diminish as they attract attention and are arbitraged away by market participants.
- Analyzing rhetorical framing in brexit debates in the house of commons the United Kingdom using large languages models: temporal trends and structural patterns-multi-stage data pipelinePublication . Beese, Julian Dieter; Shen, YufeiThis study examines rhetorical framing in House of Commons debates on the United Kingdom's exit (Brexit) from the European Union from 2012 to 2022, using Laqrge Language Models(LLMs) to analyze 62,847 parliamentary speech chunks. Building on Neuman et al.'s (1992) generic frame typology, it investigates how framing strategies have changed over time and varied across parties, referendum positions, and speaker attributes. The computational approach provides scale and precision beyond traditional qualitative methods, showing that Economic and Conflict frames dominated and that shifts in usage tracked key Brexit milestones. The findings demonstrate LLMs' value for large-scale political discourse analysis and clarify how parliamentary actors constructed Brexit's meaning during a period of constitutional turmoil
- Deep learning frameworks for enhanced user engagement predictionPublication . Guo, Zhanshuo; Shen, YufeiThis work demonstrates the individual contribution of Zhanshuo Guo in the field lab project “How Early Can We Predict Churn? Short-Window Engagement Forecasting in a Hybrid UGC Music Platform”. While chapters 1-3 summarize the collective effort of the group, chapter 4 dives into the deep learning approaches of the project. Using impression-level data from NetEase Cloud Music recently launched UGC module “Cloud Village”, this work aims at providing a scalable solution for predicting user engagement from short-term behaviors, which can help the management team to handle high user mobility issue and design win-back strategies. We employ both machine and deep learning models across two targets: binary churn and multiclass engagement. The cross design provides complementary perspectives on user behaviors, enabling performance indicators from one task to enrich the other. The result contributes to a more robust understanding of engagement prediction and operational values.
- Designing a reference KPI dashboard framework for sustainable aquaculture scale-ups: prototype development and implementation roadmap at Oceano Fresco” - fighting growth challenges in scale ups: from intuition to data driven operationPublication . Hartyani, Bendeguz Ferreyra; Soares, Bruno HortaThis project develops a reference KPI Dashboard Framework to address digitalization gaps in bivalve aquaculture scale-ups. Using Oceano Fresco as a case study, the work first establishes a theoretical foundation for data maturity and analyzes best-practice case studies (Airbnb, Legalhero, and Glovo) to identify transferable scaling strategies. The framework organizes decision-making across Executive, Operational, and Analytical layers. Beyond design, the study delivers a functional Minimum Viable Product (MVP) and a four-phase implementation roadmap. The results demonstrate how structured data architecture enables transitioning from intuition-based management to scalable, data-driven precision.