Service-oriented medical system for supporting decisions with missing and imbalanced data.

Area of Science:

• Medical Informatics
• Machine Learning
• Data Science

Background:

• Medical datasets frequently suffer from imbalanced class distributions and missing attribute values.
• These data imperfections pose significant challenges for developing accurate diagnostic decision support systems.
• Existing methods often struggle to effectively address both issues simultaneously.

Purpose of the Study:

• To propose a service-oriented support decision system (SOSDS) robust to imbalanced data and missing values.
• To develop machine learning solutions integrated as distributed Web services for medical diagnostics.
• To construct reliable decision models directly from impaired medical datasets.

Main Methods:

• Implemented a cost-sensitive support vector machine to address imbalanced data.
• Developed a novel ensemble-based approach to handle missing attribute values by splitting data into complete subspaces.
• Integrated these solutions into a distributed, service-oriented system (SOSDS).

Main Results:

• The SOSDS demonstrated robustness in handling datasets with high percentages of missing values and imbalanced class distributions.
• Evaluated the quality of SOSDS components using three ontological datasets.
• The proposed methods effectively constructed decision models from imperfect data.

Conclusions:

• The developed SOSDS offers a viable solution for diagnostic problems in the medical domain, overcoming common data challenges.
• The system's modular, service-oriented architecture facilitates integration and scalability.
• The novel ensemble approach for missing values and cost-sensitive SVM show promise for medical data analysis.