Bella Fernandez Barroso
Last Name
Fernandez Barroso
First name
Bella
Email
bella.fernandez@unisg.ch
Now showing 1 - 2 of 2
- Some of the metrics are blocked by yourconsent settings
Item type:Publication, Towards Privacy-Compliant Healthcare Data: Developing an Efficient Multilingual Filter for PII and PHI under Safe Harbor Principles(SCITEPRESS - Science and Technology Publications, 2026-10-28); Medical text anonymization is essential for data-driven healthcare but constrained by strict privacy regulations, particularly in multilingual Swiss settings. A lightweight, privacy-compliant filter for removing personally identifiable and protected health information from unstructured medical text is presented. The system combines regular expressions, rule-based logic, and locally-run guided LLMs, where local open-source models are constrained to assign a label to each word. This approach is evaluated on 50 manually annotated SwissGerman medical documents and 520 English i2b2 notes. A regex-only baseline shows substantially lower F1 score and higher leakage on both datasets. Overall, the proposed hybrid approach reduces residual identifiers while maintaining deterministic runtime and modest computational cost, enabling practical medical text anonymization in multilingual contexts.Type:conference paperJournal:Proceedings of the 18th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge ManagementVolume:2 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Automatic Price Prediction from CAD Files(SCITEPRESS - Science and Technology Publications, 2026-10-28); The paper presents a proof of concept to automate quotation generation for modular construction, thereby replacing a slow, manual process. From 975 priced STEP models, about 40 geometric and structural features were extracted to make the data suitable for machine learning. Several models were tested: while LSTM on raw STEP data failed, tree-based models (random forest and especially XGBoost) performed best, with XGBoost using six key features. The deep-learning-based TabNet underperformed due to the limited dataset size. Due to the limited number of samples for expensive designs, a two-stage pipeline is proposed: a model handles typical cases, while a classifier flags high-priced cases for manual quotation.Type:conference paperJournal:Proceedings of the 18th International Joint Conference on Knowledge Discovery, Knowledge Engineering and Knowledge ManagementVolume:1