Adriano Galindo Leal

Aiming to reduce manual room naming in BIM projects, this study analyzed the feasibility of automatic room classification in Brazilian single-family dwellings via Graph Neural Networks (GNNs). Due to real-world data scarcity, 200 synthetic single-story house models were generated and converted into graphs capturing geometric attributes and spatial relationships. The SAGE-E model was tested under various combinations of hidden layers, nodes, activation functions, learning rates, batch sizes, and aggregators, achieving a 0.96 accuracy on the test set. Despite this high performance, limitations occurred when differentiating semantically similar rooms, reflecting both model constraints and the complexity of Brazilian residential architecture. In conclusion, this approach can reduce repetitive tasks in BIM workflows, a potential that could be enhanced by expanding the database and incorporating additional semantic attributes.

Accurate prediction of posttransplant survival is critical for optimizing liver allocation under organ scarcity. Although the Model for End-Stage Liver Disease (MELD) is widely used for prioritization, it was not designed to estimate posttransplant outcomes, limiting its utility for individualized risk stratification. We developed and temporally validated machine-learning models to predict 12-month survival after liver transplantation using routinely collected donor–recipient variables from a real-world cohort. Sequential temporal validation was used to approximate prospective deployment. Performance was evaluated in terms of discrimination (receiver operating characteristic–area under the curve [ROC-AUC] and precision–recall area under the curve), calibration (Brier score), and decision-analytic evaluation using decision curve analysis. Class imbalance was addressed through cost-sensitive learning, and predicted probabilities were recalibrated with Platt scaling. Across temporal validation windows, discrimination remained modest, with ROC-AUC values below 0.70 and reaching 0.599 in the final window, reflecting the complexity of posttransplant outcomes in heterogeneous clinical populations. Calibration remained relatively stable (Brier score ≈ 0.18). Within the evaluated threshold range, the machine-learning models yielded a positive net benefit and generally exceeded the transformed MELD baseline in decision-analytic comparisons. However, these findings should be interpreted cautiously given the modest discrimination and the exploratory nature of the MELD transformation used for cross-model comparison. Logistic regression provided the most consistent balance of discrimination, calibration, and net benefit. Overall, calibrated models with modest discrimination may still offer complementary risk information. However, further external validation, threshold-specific evaluation, and prospective testing are required before such models can be considered for operational decision support.

The increasing demand for proactive patient out-reach and follow-up within health insurance providers and large hospital systems has led to the adoption of automated message classification systems. Although state-of-the-art tools like Azure and OpenAI o1 series are available, their costs are prohibitive for small companies. Therefore, using cost-effective traditional tools for text classification is still a valuable alternative. This study investigates how OpenAI's ChatGPT can generate synthetic data to balance class distributions in medical message datasets and improve the performance of binary classification models. The primary aim is to classify messages that indicate medical needs, aiding in patient outreach and efficient resource allocation. We applied Natural Language Processing (NLP) techniques, including Logistic Regression, Naive Bayes, Random Forest, XGBoost, and BERTimbau, a Portuguese transformer-based model. Models were trained on both original and augmented datasets, and results showed a significant increase in classification accuracy, particularly with BERTimbau, whose performance improved from 84% to 99%. Synthetic data also reduced overfitting by narrowing the gap between training and validation accuracies while improving model robustness on unseen data. Our findings demonstrate that synthetic data from ChatGPT effectively addresses class imbalance, improving the robustness and accuracy of medical message classification models. However, challenges remain in correctly classifying non-medical messages, indicating that further refinement in synthetic data generation is necessary for optimal performance in real-world scenarios. These results suggest that such approaches can significantly enhance automated patient engagement efforts.

We introduce Pantheon-DNA, an end-to-end processing pipeline for DNA data storage that effectively addresses scalability challenges while efficiently managing large datasets, maintaining ≥99.996% retrievability at 10× coverage under both LER and HER in our tests. To prevent repetitive patterns in DNA sequences, which potentially cause chimeras at the molecular level and also hinder clustering algorithms, we propose a data arrangement scheme and a randomization procedure during encoding. We use block data architecture to enhance parallel processing and retrieval. The proposed sequencing data preprocessing pipeline utilizes prior knowledge of the data structure encoded in the DNA sequences to simplify conventional clustering routines and reduce computational complexity. The system's robustness and reliability are validated through an actual synthesis and sequencing experiment, which encodes and decodes 1.59 MB of data containing multiple files. Future enhancements will focus on refining error correction capabilities, particularly for indel recovery, as well as optimizing preprocessing efficiency and sensitivity.

Recent advancements in anti-money laundering (AML) strategies underscore the imperative for more accurate and efficient detection systems. This paper delineates a robust approach utilising machine learning (ML) and artificial intelligence (AI) to refine AML frameworks within national financial institutions, reducing false positive alerts. Integrating data from ‘Know Your Customer’ (KYC) initiatives with comprehensive transactional datasets, our methodology identifies high-risk transactions and complies with rigorous regulatory standards while curtailing operational costs. Through a comparative analysis of various ML models—including Decision Trees, Random Forest, Support Vector Machines, Logistic Regression, Artificial Neural Networks (ANN), and Convolutional Neural Networks (CNN) —we have identified that the Random Forest model notably decreased false positive rates to 2.1% and maintained an elevated detection rate for potentially illicit transactions. This model markedly surpasses traditional rule-based systems in performance, confirming its efficacy and suitability for broad implementation. By streamlining AML processes and diminishing compliance-related expenditures, this study presents a scalable and efficient model for financial institutions, optimising operational efficiency and fostering better cost-effectiveness in combating economic crimes. This analysis identifies the most effective models that balance detection accuracy with the need for explainability, an essential requirement for gaining trust from regulatory bodies and internal compliance teams. The findings of this study contribute to the field by offering a replicable and efficient AML framework that can safeguard financial institutions against regulatory penalties and reputational risks.