RELEAP: Reinforcement-Enhanced Label-Efficient Active Phenotyping for Electronic Health Records

arXiv — cs.LG•Wednesday, November 12, 2025 at 5:00:00 AM

RELEAP, a new framework for electronic health records, leverages reinforcement learning to enhance active phenotyping, particularly in predicting lung cancer risk. By integrating feedback from downstream prediction models, RELEAP adapts its querying strategies, leading to improved accuracy over traditional methods. Evaluated on a cohort from Duke University Health System, it demonstrated notable performance gains, with an AUC increase from 0.774 to 0.805 and a C-index rise from 0.718 to 0.752. These improvements are significant as they tackle the issue of unreliable proxy labels in health data, which can compromise risk predictions. The ability to refine phenotypes based on actual predictive performance marks a pivotal shift in how health data can be utilized, promising enhanced patient care and outcomes through more accurate risk assessments.

— via World Pulse Now AI Editorial System

Read Original

Was this article worth reading? Share it

Recommended Readings

arXiv — cs.CV2 days ago

Contrastive Integrated Gradients: A Feature Attribution-Based Method for Explaining Whole Slide Image Classification

PositiveArtificial Intelligence

The paper introduces Contrastive Integrated Gradients (CIG), a new method for enhancing interpretability in Whole Slide Image (WSI) analysis within computational pathology. CIG addresses challenges posed by high-resolution images, improving the identification of class-discriminative signals crucial for tumor subtype differentiation. By computing contrastive gradients in logit space, CIG provides clearer distinctions between tumor and non-tumor areas, ensuring consistency and theoretical soundness in attribution.

Read full article

via arXiv — cs.CV

arXiv — cs.CL2 days ago

Automated Analysis of Learning Outcomes and Exam Questions Based on Bloom's Taxonomy

NeutralArtificial Intelligence

This paper investigates the automated classification of exam questions and learning outcomes based on Bloom's Taxonomy. A dataset of 600 sentences was categorized into six cognitive levels: Knowledge, Comprehension, Application, Analysis, Synthesis, and Evaluation. Various machine learning models, including traditional methods and large language models, were evaluated, with Support Vector Machines achieving the highest accuracy of 94%, while RNN models and BERT faced significant overfitting issues.

Read full article

via arXiv — cs.CL

arXiv — cs.LG2 days ago

How Data Quality Affects Machine Learning Models for Credit Risk Assessment

PositiveArtificial Intelligence

Machine Learning (ML) models are increasingly used for credit risk evaluation, with their effectiveness dependent on data quality. This research investigates the impact of data quality issues such as missing values, noisy attributes, outliers, and label errors on the predictive accuracy of ML models. Using an open-source dataset, the study assesses the robustness of ten commonly used models, including Random Forest, SVM, and Logistic Regression, revealing significant differences in model performance based on data degradation.

Read full article

via arXiv — cs.LG

arXiv — cs.LG2 days ago

Toward Scalable Early Cancer Detection: Evaluating EHR-Based Predictive Models Against Traditional Screening Criteria

PositiveArtificial Intelligence

Current cancer screening guidelines are limited to a few cancer types and rely on specific criteria like age or smoking history to identify high-risk individuals. A study evaluates the effectiveness of predictive models using electronic health records (EHRs) to identify high-risk groups by detecting subtle prediagnostic signals of cancer. The research focuses on eight major cancers, including breast and lung cancer, and compares EHR-based models to traditional risk factors. Evidence suggests EHR-based models may be more effective in identifying true cancer cases.

Read full article

via arXiv — cs.LG