Clinical Decision Support Systems: From the Perspective of Small and Imbalanced Data Set

Area of Science:

• Health Informatics
• Machine Learning
• Data Mining

Background:

• Clinical decision support systems (CDSS) are crucial for healthcare, aiding professionals in patient data analysis.
• Designing effective CDSS is challenged by the complexity of diseases and the nature of health data, which is often small and imbalanced.
• Traditional classification algorithms struggle with imbalanced datasets, leading to misclassification of minority classes.

Purpose of the Study:

• To investigate methods for improving classification performance on small and imbalanced health datasets.
• To evaluate the effectiveness of distance-based data generation methods for oversampling.
• To recommend an optimal synthetic data generation rate for enhanced diagnostic accuracy in CDSS.

Main Methods:

• Utilized a publicly accessible hepatitis dataset.
• Applied distance-based data generation techniques for oversampling the imbalanced dataset.
• Classified the oversampled data using four machine learning algorithms: Artificial Neural Networks, Support Vector Machines, Naive Bayes, and Decision Tree.

Main Results:

• Oversampling techniques demonstrated potential in mitigating the challenges posed by small and imbalanced datasets.
• Comparative analysis of four machine learning algorithms provided insights into their performance on synthetic data.
• Classification scores indicated varying effectiveness of algorithms based on the synthetic data generation rate.

Conclusions:

• Distance-based oversampling can enhance the performance of machine learning algorithms in health data analysis.
• The study recommends an optimal synthetic data generation rate based on classification scores for improved CDSS.
• Addressing data imbalance is critical for advancing the reliability and accuracy of clinical decision support systems.