https://scholar.google.com/citations?user=DxZvvGUAAAAJ&hl=en
Title: A comparative study of regularization techniques in machine learning for community-based health screening tools
Abstract: This project explores the role of regularization techniques in enhancing the performance and generalizability of machine learning models for health screening in low-resource, community-based settings. Using osteoporosis as a case study, the project investigates how different regularization strategies, such as L1 (Lasso), L2 (Ridge), Elastic Net, and tree-based pruning, impact model behavior when applied to small-scale, imbalanced clinical datasets. Emphasis is placed on the mathematical foundations of regularization, including bias–variance trade-offs and penalty-based optimization. By comparing models such as logistic regression, decision trees, and ensemble methods (e.g., XGBoost), the study aims to identify the optimal regularization configurations that balance predictive performance, model interpretability, and feasibility for field-level deployment. The project further highlights the importance of simplicity and robustness in designing screening tools that can be adopted by non-specialist health workers.
Prerequisites:
· Basic understanding of supervised machine learning models (e.g., logistic regression, decision trees),
· Familiarity with regularization methods (e.g., Lasso, Ridge, pruning),
· Competence in linear algebra and optimization principles,
· Interest in health data analysis and practical algorithm design.
Group members
KITTISAK THAWNASHOM
RESA MAE SANGCO
Wannapa PANITSUPAKAMON
Cielomi JALON
SITTHIPON PHUMSAMUD
Busadee PRATUMVINIT